c2be8cb · crush · git.iko.soy

git.iko.soy/crush

on main help sign in

fix: set non-interactive env vars in shell to prevent editor hangs (#3025)

c2be8cbTai Groot5 months ago

internal/agent/testdata/TestCoderAgent/glm-5.1/bash_tool.yaml+459−525

compare with main

unifiedsplit

@@ -6,9 +6,9 @@ interactions:

6 proto: HTTP/1.1

7 proto_major: 1

8 proto_minor: 1

9− content_length: 853

9+ content_length: 33985

10 host: ""

11− body: '{"messages":[{"content":"You will generate a short title based on the first message a user begins a conversation with.\n\n<rules>\n- Keep the title in the same language that the user wrote their message in.\n- Ensure it is not more than 50 characters long.\n- The title should be a summary of the user''s message.\n- It should be one line long.\n- Do not use quotes or colons.\n- The entire text you return will be used as the title.\n- Never return anything that is more than one sentence (one line) long.\n</rules>\n\n /no_think","role":"system"},{"content":"Generate a concise title for the following content:\n\nuse bash to create a file named test.txt with content ''hello bash''. do not print its timestamp\n <think>\n\n</think>","role":"user"}],"model":"gpt-oss-120b","max_completion_tokens":40,"stream_options":{"include_usage":true},"stream":true}'

11+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/bash_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use bash to create a file named test.txt with content 'hello bash'. do not print its timestamp\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

12 headers:

13 Accept:

14 - application/json

@@ -25,927 +25,871 @@ interactions:

25 content_length: -1

26 uncompressed: true

27 body: |+

28− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

28+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

29−

30− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"We"},"finish_reason":null}]}

31−

32− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" need"},"finish_reason":null}]}

33−

34− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

35−

36− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" generate"},"finish_reason":null}]}

37−

38− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

39−

40− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

41−

42− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

43−

44− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

45−

46− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" same"},"finish_reason":null}]}

47−

48− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" language"},"finish_reason":null}]}

49−

50− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" as"},"finish_reason":null}]}

51−

52− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

53−

54− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

55−

56− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

57−

58− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"English"},"finish_reason":null}]}

59−

60− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")."},"finish_reason":null}]}

61−

62− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Summ"},"finish_reason":null}]}

63−

64− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ar"},"finish_reason":null}]}

65−

66− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ize"},"finish_reason":null}]}

67−

68− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

69−

70− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

71−

72− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Create"},"finish_reason":null}]}

74− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" test"},"finish_reason":null}]}

30+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

76− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

32+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

78− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

34+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" wants"},"finish_reason":null}]}

80− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" hello"},"finish_reason":null}]}

36+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" me"},"finish_reason":null}]}

82− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" bash"},"finish_reason":null}]}

38+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

84− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" using"},"finish_reason":null}]}

40+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" create"},"finish_reason":null}]}

86− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" bash"},"finish_reason":null}]}

42+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

88− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

44+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" file"},"finish_reason":null}]}

90− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

46+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" named"},"finish_reason":null}]}

92− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" timestamp"},"finish_reason":null}]}

48+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" test"},"finish_reason":null}]}

94− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

50+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

96− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Title"},"finish_reason":null}]}

52+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

98− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \u003c="},"finish_reason":null}]}

54+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

100− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

56+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" '"},"finish_reason":null}]}

101

102− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

58+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"hello"},"finish_reason":null}]}

103

104− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

60+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" bash"},"finish_reason":null}]}

105

106− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

62+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"'"},"finish_reason":null}]}

107

108− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" one"},"finish_reason":null}]}

64+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" using"},"finish_reason":null}]}

109

110− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

66+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

111

112− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

68+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" bash"},"finish_reason":null}]}

113

114− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

70+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" tool"},"finish_reason":null}]}

115

116− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

72+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

117

118− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" or"},"finish_reason":null}]}

74+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

119

120− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" col"},"finish_reason":null}]}

76+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" not"},"finish_reason":null}]}

121

122− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ons"},"finish_reason":null}]}

78+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" print"},"finish_reason":null}]}

123

124− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

80+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" its"},"finish_reason":null}]}

125

126− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" So"},"finish_reason":null}]}

82+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" timestamp"},"finish_reason":null}]}

127

128− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" something"},"finish_reason":null}]}

84+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

129

130− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" like"},"finish_reason":null}]}

86+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_maHr9QegPpRMRiFEH9e4bSTo","type":"function","index":0,"function":{"name":"bash","arguments":""}}]},"finish_reason":null}]}

131

132− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

88+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

133

134− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Bash"},"finish_reason":null}]}

90+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

135

136− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" command"},"finish_reason":null}]}

92+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"command\": "}}]},"finish_reason":null}]}

137

138− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

94+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"echo"}}]},"finish_reason":null}]}

139

140− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" create"},"finish_reason":null}]}

96+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" '"}}]},"finish_reason":null}]}

141

142− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" test"},"finish_reason":null}]}

98+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"hello"}}]},"finish_reason":null}]}

143

144− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

100+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" bash"}}]},"finish_reason":null}]}

145

101

146− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

102+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"'"}}]},"finish_reason":null}]}

147

103

148− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" hello"},"finish_reason":null}]}

104+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" \u003e"}}]},"finish_reason":null}]}

149

105

150− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" bash"},"finish_reason":null}]}

106+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" test"}}]},"finish_reason":null}]}

151

107

152− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" without"},"finish_reason":null}]}

108+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".txt"}}]},"finish_reason":null}]}

153

109

154− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" timestamp"},"finish_reason":null}]}

110+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\", \"description\": "}}]},"finish_reason":null}]}

155

111

156− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

112+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"Create"}}]},"finish_reason":null}]}

157

113

158− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

114+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" test"}}]},"finish_reason":null}]}

159

115

160− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

116+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".txt"}}]},"finish_reason":null}]}

161

117

162− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

118+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" with"}}]},"finish_reason":null}]}

163

119

164− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

120+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" content"}}]},"finish_reason":null}]}

165

121

166− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"B"},"finish_reason":null}]}

122+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"}"}}]},"finish_reason":null}]}

167

123

168− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ash"},"finish_reason":null}]}

124+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

169

125

170− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" command"},"finish_reason":null}]}

126+ data: {"id":"chatcmpl-3e35182185e98776cee50f17","object":"chat.completion.chunk","created":1780069679,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":715,"completion_tokens":56,"total_tokens":7875,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":32},"cost":{"usd":0.0013062500000000001,"hypercredits":0.026125000000000002},"remaining":{"hypercredits":0}}}

171

127

172− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

128+ data: [DONE]

173−

174− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" create"},"finish_reason":null}]}

175−

176− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" test"},"finish_reason":null}]}

177−

178− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

179−

180− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

181−

182− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" hello"},"finish_reason":null}]}

183−

184− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" bash"},"finish_reason":null}]}

185

129

130+ headers:

131+ Content-Type:

132+ - text/event-stream

133+ status: 200 OK

134+ code: 200

135+ duration: 568.683583ms

136+- id: 1

137+ request:

138+ proto: HTTP/1.1

139+ proto_major: 1

140+ proto_minor: 1

141+ content_length: 853

142+ host: ""

143+ body: '{"messages":[{"content":"You will generate a short title based on the first message a user begins a conversation with.\n\n<rules>\n- Keep the title in the same language that the user wrote their message in.\n- Ensure it is not more than 50 characters long.\n- The title should be a summary of the user''s message.\n- It should be one line long.\n- Do not use quotes or colons.\n- The entire text you return will be used as the title.\n- Never return anything that is more than one sentence (one line) long.\n</rules>\n\n /no_think","role":"system"},{"content":"Generate a concise title for the following content:\n\nuse bash to create a file named test.txt with content ''hello bash''. do not print its timestamp\n <think>\n\n</think>","role":"user"}],"model":"gpt-oss-120b","max_completion_tokens":40,"stream_options":{"include_usage":true},"stream":true}'

144+ headers:

145+ Accept:

146+ - application/json

147+ Content-Type:

148+ - application/json

149+ User-Agent:

150+ - Charm-Crush/devel (https://charm.land/crush)

151+ url: https://hyper.charm.land/v1/chat/completions

152+ method: POST

153+ response:

154+ proto: HTTP/2.0

155+ proto_major: 2

156+ proto_minor: 0

157+ content_length: -1

158+ uncompressed: true

159+ body: |+

186− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" without"},"finish_reason":null}]}

160+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

187

161

188− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" timestamp"},"finish_reason":null}]}

162+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"We"},"finish_reason":null}]}

189

163

190− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

164+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" need"},"finish_reason":null}]}

191

165

192− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" length"},"finish_reason":null}]}

166+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

193

167

194− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"?"},"finish_reason":null}]}

168+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" produce"},"finish_reason":null}]}

195

169

196− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Let's"},"finish_reason":null}]}

170+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

197

171

198− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

172+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" short"},"finish_reason":null}]}

199

173

200− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

174+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

201

175

202− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" B"},"finish_reason":null}]}

176+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

203

177

204− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

178+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"max"},"finish_reason":null}]}

205

179

206− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

180+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

207

181

208− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"),"},"finish_reason":null}]}

182+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

209

183

210− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

184+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

211

185

212− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"2"},"finish_reason":null}]}

186+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

213

187

214− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

188+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

215

189

216− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" s"},"finish_reason":null}]}

190+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" same"},"finish_reason":null}]}

217

191

218− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"3"},"finish_reason":null}]}

192+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" language"},"finish_reason":null}]}

219

193

220− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

194+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

221

195

222− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" h"},"finish_reason":null}]}

196+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"English"},"finish_reason":null}]}

223

197

224− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

198+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")."},"finish_reason":null}]}

225

199

226− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

200+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Summ"},"finish_reason":null}]}

227

201

228− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

202+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ar"},"finish_reason":null}]}

229

203

230− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"5"},"finish_reason":null}]}

204+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ize"},"finish_reason":null}]}

231

205

232− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

206+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

233

207

234− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" c"},"finish_reason":null}]}

208+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

235

209

236− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"6"},"finish_reason":null}]}

210+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"use"},"finish_reason":null}]}

237

211

238− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",o"},"finish_reason":null}]}

212+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" bash"},"finish_reason":null}]}

239

213

240− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"7"},"finish_reason":null}]}

214+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

241

215

242− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",m"},"finish_reason":null}]}

216+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" create"},"finish_reason":null}]}

243

217

244− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"8"},"finish_reason":null}]}

218+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

245

219

246− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",m"},"finish_reason":null}]}

220+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" file"},"finish_reason":null}]}

247

221

248− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"9"},"finish_reason":null}]}

222+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" named"},"finish_reason":null}]}

249

223

250− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",a"},"finish_reason":null}]}

224+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" test"},"finish_reason":null}]}

251

225

252− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"10"},"finish_reason":null}]}

226+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

253

227

254− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",n"},"finish_reason":null}]}

228+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

255

229

256− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"11"},"finish_reason":null}]}

230+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

257

231

258− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",d"},"finish_reason":null}]}

232+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" '"},"finish_reason":null}]}

259

233

260− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"12"},"finish_reason":null}]}

234+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"hello"},"finish_reason":null}]}

261

235

262− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

236+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" bash"},"finish_reason":null}]}

263

237

264− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

238+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"'."},"finish_reason":null}]}

265

239

266− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"13"},"finish_reason":null}]}

240+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" do"},"finish_reason":null}]}

267

241

268− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

242+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" not"},"finish_reason":null}]}

269

243

270− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

244+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" print"},"finish_reason":null}]}

271

245

272− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"14"},"finish_reason":null}]}

246+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" its"},"finish_reason":null}]}

273

247

274− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",o"},"finish_reason":null}]}

248+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" timestamp"},"finish_reason":null}]}

275

249

276− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"15"},"finish_reason":null}]}

250+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

277

251

278− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

252+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Title"},"finish_reason":null}]}

279

253

280− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

254+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" could"},"finish_reason":null}]}

281

255

282− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"16"},"finish_reason":null}]}

256+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

283

257

284− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",c"},"finish_reason":null}]}

258+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

285

259

286− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"17"},"finish_reason":null}]}

260+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Create"},"finish_reason":null}]}

287

261

288− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",r"},"finish_reason":null}]}

262+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" test"},"finish_reason":null}]}

289

263

290− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"18"},"finish_reason":null}]}

264+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

291

265

292− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",e"},"finish_reason":null}]}

266+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

293

267

294− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"19"},"finish_reason":null}]}

268+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" hello"},"finish_reason":null}]}

295

269

296− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",a"},"finish_reason":null}]}

270+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" bash"},"finish_reason":null}]}

297

271

298− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"20"},"finish_reason":null}]}

272+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" via"},"finish_reason":null}]}

299

273

300− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",t"},"finish_reason":null}]}

274+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" bash"},"finish_reason":null}]}

301

275

302− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"21"},"finish_reason":null}]}

276+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" without"},"finish_reason":null}]}

303

277

304− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",e"},"finish_reason":null}]}

278+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" timestamp"},"finish_reason":null}]}

305

279

306− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"22"},"finish_reason":null}]}

280+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

307

281

308− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

282+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

309

283

310− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

284+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

311

285

312− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"23"},"finish_reason":null}]}

286+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

313

287

314− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",t"},"finish_reason":null}]}

288+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

315

289

316− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"24"},"finish_reason":null}]}

290+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Create"},"finish_reason":null}]}

317

291

318− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",e"},"finish_reason":null}]}

292+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" test"},"finish_reason":null}]}

319

293

320− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"25"},"finish_reason":null}]}

294+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

321

295

322− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",s"},"finish_reason":null}]}

296+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

323

297

324− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"26"},"finish_reason":null}]}

298+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" hello"},"finish_reason":null}]}

325

299

326− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",t"},"finish_reason":null}]}

300+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" bash"},"finish_reason":null}]}

327

301

328− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"27"},"finish_reason":null}]}

302+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" via"},"finish_reason":null}]}

329

303

330− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",."},"finish_reason":null}]}

304+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" bash"},"finish_reason":null}]}

331

305

332− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"28"},"finish_reason":null}]}

306+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" without"},"finish_reason":null}]}

333

307

334− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",t"},"finish_reason":null}]}

308+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" timestamp"},"finish_reason":null}]}

335

309

336− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"29"},"finish_reason":null}]}

310+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

337

311

338− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",x"},"finish_reason":null}]}

312+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Let's"},"finish_reason":null}]}

339

313

340− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"30"},"finish_reason":null}]}

314+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

341

315

342− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",t"},"finish_reason":null}]}

316+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

343

317

344− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"31"},"finish_reason":null}]}

318+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

345

319

346− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

320+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Create"},"finish_reason":null}]}

347

321

348− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

322+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" test"},"finish_reason":null}]}

349

323

350− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"32"},"finish_reason":null}]}

324+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

351

325

352− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",w"},"finish_reason":null}]}

326+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

353

327

354− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"33"},"finish_reason":null}]}

328+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" hello"},"finish_reason":null}]}

355

329

356− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",i"},"finish_reason":null}]}

330+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" bash"},"finish_reason":null}]}

357

331

358− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"34"},"finish_reason":null}]}

332+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" via"},"finish_reason":null}]}

359

333

360− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",t"},"finish_reason":null}]}

334+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" bash"},"finish_reason":null}]}

361

335

362− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"35"},"finish_reason":null}]}

336+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" without"},"finish_reason":null}]}

363

337

364− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",h"},"finish_reason":null}]}

338+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" timestamp"},"finish_reason":null}]}

365

339

366− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"36"},"finish_reason":null}]}

340+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

367

341

368− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

342+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" length"},"finish_reason":null}]}

369

343

370− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

344+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

371

345

372− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"37"},"finish_reason":null}]}

346+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

373

347

374− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",h"},"finish_reason":null}]}

348+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

375

349

376− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"38"},"finish_reason":null}]}

350+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

377

351

378− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",e"},"finish_reason":null}]}

352+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" C"},"finish_reason":null}]}

379

353

380− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"39"},"finish_reason":null}]}

354+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

381

355

382− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",l"},"finish_reason":null}]}

356+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

383

357

384− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"40"},"finish_reason":null}]}

358+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

385

359

386− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",l"},"finish_reason":null}]}

360+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" r"},"finish_reason":null}]}

387

361

388− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"41"},"finish_reason":null}]}

362+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"2"},"finish_reason":null}]}

389

363

390− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",o"},"finish_reason":null}]}

364+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

391

365

392− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"42"},"finish_reason":null}]}

366+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"3"},"finish_reason":null}]}

393

367

394− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

368+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

395

369

396− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

370+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

397

371

398− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"43"},"finish_reason":null}]}

372+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

399

373

400− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",b"},"finish_reason":null}]}

374+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"5"},"finish_reason":null}]}

401

375

402− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"44"},"finish_reason":null}]}

376+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

403

377

404− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",a"},"finish_reason":null}]}

378+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"6"},"finish_reason":null}]}

405

379

406− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"45"},"finish_reason":null}]}

380+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

407

381

408− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",s"},"finish_reason":null}]}

382+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"7"},"finish_reason":null}]}

409

383

410− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"46"},"finish_reason":null}]}

384+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

411

385

412− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",h"},"finish_reason":null}]}

386+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"8"},"finish_reason":null}]}

413

387

414− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"47"},"finish_reason":null}]}

388+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

415

389

416− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

390+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"9"},"finish_reason":null}]}

417

391

418− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

392+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" s"},"finish_reason":null}]}

419

393

420− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"48"},"finish_reason":null}]}

394+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"10"},"finish_reason":null}]}

421

395

422− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",w"},"finish_reason":null}]}

396+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

423

397

424− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"49"},"finish_reason":null}]}

398+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"11"},"finish_reason":null}]}

425

399

426− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",i"},"finish_reason":null}]}

400+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

427

401

428− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

402+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"12"},"finish_reason":null}]}

429

403

430− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",t"},"finish_reason":null}]}

404+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

431

405

432− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"51"},"finish_reason":null}]}

406+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"13"},"finish_reason":null}]}

433

407

434− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",h"},"finish_reason":null}]}

408+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" x"},"finish_reason":null}]}

435

409

436− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"52"},"finish_reason":null}]}

410+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"14"},"finish_reason":null}]}

437

411

438− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",o"},"finish_reason":null}]}

412+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

439

413

440− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"53"},"finish_reason":null}]}

414+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"15"},"finish_reason":null}]}

441

415

442− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",u"},"finish_reason":null}]}

416+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

443

417

444− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"54"},"finish_reason":null}]}

418+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"16"},"finish_reason":null}]}

445

419

446− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",t"},"finish_reason":null}]}

420+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" w"},"finish_reason":null}]}

447

421

448− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"55"},"finish_reason":null}]}

422+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"17"},"finish_reason":null}]}

449

423

450− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

424+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" i"},"finish_reason":null}]}

451

425

452− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

426+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"18"},"finish_reason":null}]}

453

427

454− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"56"},"finish_reason":null}]}

428+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

455

429

456− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",t"},"finish_reason":null}]}

430+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"19"},"finish_reason":null}]}

457

431

458− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"57"},"finish_reason":null}]}

432+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" h"},"finish_reason":null}]}

459

433

460− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",i"},"finish_reason":null}]}

434+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"20"},"finish_reason":null}]}

461

435

462− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"58"},"finish_reason":null}]}

436+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

463

437

464− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",m"},"finish_reason":null}]}

438+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"21"},"finish_reason":null}]}

465

439

466− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"59"},"finish_reason":null}]}

440+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" h"},"finish_reason":null}]}

467

441

468− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",e"},"finish_reason":null}]}

442+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"22"},"finish_reason":null}]}

469

443

470− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"60"},"finish_reason":null}]}

444+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

471

445

472− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",s"},"finish_reason":null}]}

446+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"23"},"finish_reason":null}]}

473

447

474− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"61"},"finish_reason":null}]}

448+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" l"},"finish_reason":null}]}

475

449

476− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",t"},"finish_reason":null}]}

450+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"24"},"finish_reason":null}]}

477

451

478− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"62"},"finish_reason":null}]}

452+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" l"},"finish_reason":null}]}

479

453

480− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",a"},"finish_reason":null}]}

454+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"25"},"finish_reason":null}]}

481

455

482− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"63"},"finish_reason":null}]}

456+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" o"},"finish_reason":null}]}

483

457

484− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",m"},"finish_reason":null}]}

458+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"26"},"finish_reason":null}]}

485

459

486− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"64"},"finish_reason":null}]}

460+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

487

461

488− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":",p"},"finish_reason":null}]}

462+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"27"},"finish_reason":null}]}

489

463

490− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"65"},"finish_reason":null}]}

464+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" b"},"finish_reason":null}]}

491

465

492− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

466+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"28"},"finish_reason":null}]}

493

467

494− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

468+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

495

469

496− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \u003e"},"finish_reason":null}]}

470+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"29"},"finish_reason":null}]}

497

471

498− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

472+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" s"},"finish_reason":null}]}

499

473

500− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

474+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"30"},"finish_reason":null}]}

501

475

502− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" We"},"finish_reason":null}]}

476+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" h"},"finish_reason":null}]}

503

477

504− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" need"},"finish_reason":null}]}

478+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"31"},"finish_reason":null}]}

505

479

506− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \u003c="},"finish_reason":null}]}

480+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

507

481

508− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

482+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"32"},"finish_reason":null}]}

509

483

510− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

484+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" v"},"finish_reason":null}]}

511

485

512− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Let's"},"finish_reason":null}]}

486+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"33"},"finish_reason":null}]}

513

487

514− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" shorten"},"finish_reason":null}]}

488+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" i"},"finish_reason":null}]}

515

489

516− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".\n\n"},"finish_reason":null}]}

490+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"34"},"finish_reason":null}]}

517

491

518− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Maybe"},"finish_reason":null}]}

492+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

519

493

520− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

494+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"35"},"finish_reason":null}]}

521

495

522− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Bash"},"finish_reason":null}]}

496+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

523

497

524− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" create"},"finish_reason":null}]}

498+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"36"},"finish_reason":null}]}

525

499

526− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" test"},"finish_reason":null}]}

500+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" b"},"finish_reason":null}]}

527

501

528− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

502+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"37"},"finish_reason":null}]}

529

503

530− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

504+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

531

505

532− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

506+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"38"},"finish_reason":null}]}

533

507

534− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"hello"},"finish_reason":null}]}

508+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" s"},"finish_reason":null}]}

535

509

536− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" bash"},"finish_reason":null}]}

510+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"39"},"finish_reason":null}]}

537

511

538− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

512+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" h"},"finish_reason":null}]}

539

513

540− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

514+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"40"},"finish_reason":null}]}

541

515

542− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" timestamp"},"finish_reason":null}]}

516+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

543

517

544− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

518+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"41"},"finish_reason":null}]}

545

519

546− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

520+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" w"},"finish_reason":null}]}

547

521

548− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

522+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"42"},"finish_reason":null}]}

549

523

550− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

524+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" i"},"finish_reason":null}]}

551

525

552− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"B"},"finish_reason":null}]}

526+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"43"},"finish_reason":null}]}

553

527

554− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ash"},"finish_reason":null}]}

528+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

555

529

556− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" create"},"finish_reason":null}]}

530+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"44"},"finish_reason":null}]}

557

531

558− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" test"},"finish_reason":null}]}

532+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" h"},"finish_reason":null}]}

559

533

560− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

534+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"45"},"finish_reason":null}]}

561

535

562− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

536+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" o"},"finish_reason":null}]}

563

537

564− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" hello"},"finish_reason":null}]}

538+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"46"},"finish_reason":null}]}

565

539

566− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" bash"},"finish_reason":null}]}

540+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" u"},"finish_reason":null}]}

567

541

568− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

542+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"47"},"finish_reason":null}]}

569

543

570− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" timestamp"},"finish_reason":null}]}

544+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

571

545

572− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

546+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"48"},"finish_reason":null}]}

573

547

574− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

548+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

575

549

576− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

550+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"49"},"finish_reason":null}]}

577

551

578− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

552+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

579

553

580− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" B"},"finish_reason":null}]}

554+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

581

555

582− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

556+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" i"},"finish_reason":null}]}

583

557

584− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

558+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"51"},"finish_reason":null}]}

585

559

586− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"2"},"finish_reason":null}]}

560+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" m"},"finish_reason":null}]}

587

561

588− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" s"},"finish_reason":null}]}

562+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"52"},"finish_reason":null}]}

589

563

590− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"3"},"finish_reason":null}]}

564+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

591

565

592− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" h"},"finish_reason":null}]}

566+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"53"},"finish_reason":null}]}

593

567

594− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

568+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" s"},"finish_reason":null}]}

595

569

596− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

570+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"54"},"finish_reason":null}]}

597

571

598− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"5"},"finish_reason":null}]}

572+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

599

573

600− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" c"},"finish_reason":null}]}

574+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"55"},"finish_reason":null}]}

601

575

602− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"6"},"finish_reason":null}]}

576+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

603

577

604− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" r"},"finish_reason":null}]}

578+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"56"},"finish_reason":null}]}

605

579

606− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"7"},"finish_reason":null}]}

580+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" m"},"finish_reason":null}]}

607

581

608− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

582+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"57"},"finish_reason":null}]}

609

583

610− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"8"},"finish_reason":null}]}

584+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" p"},"finish_reason":null}]}

611

585

612− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

586+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"58"},"finish_reason":null}]}

613

587

614− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"9"},"finish_reason":null}]}

588+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

615

589

616− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

590+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

617

591

618− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"10"},"finish_reason":null}]}

592+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

619

593

620− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

594+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"58"},"finish_reason":null}]}

621

595

622− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"11"},"finish_reason":null}]}

596+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \u003e"},"finish_reason":null}]}

623

597

624− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

598+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

625

599

626− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"12"},"finish_reason":null}]}

600+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

627

601

628− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

602+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Need"},"finish_reason":null}]}

629

603

630− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"13"},"finish_reason":null}]}

604+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \u003c="},"finish_reason":null}]}

631

605

632− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

606+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

633

607

634− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"14"},"finish_reason":null}]}

608+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".\n\n"},"finish_reason":null}]}

635

609

636− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" s"},"finish_reason":null}]}

610+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Try"},"finish_reason":null}]}

637

611

638− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"15"},"finish_reason":null}]}

612+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" shorter"},"finish_reason":null}]}

639

613

640− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

614+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

641

615

642− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"16"},"finish_reason":null}]}

616+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

643

617

644− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

618+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Create"},"finish_reason":null}]}

645

619

646− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"17"},"finish_reason":null}]}

620+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" test"},"finish_reason":null}]}

647

621

648− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

622+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

649

623

650− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"18"},"finish_reason":null}]}

624+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

651

625

652− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" x"},"finish_reason":null}]}

626+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" hello"},"finish_reason":null}]}

653

627

654− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"19"},"finish_reason":null}]}

628+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" bash"},"finish_reason":null}]}

655

629

656− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

630+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

657

631

658− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"20"},"finish_reason":null}]}

632+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

659

633

660− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

634+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" timestamp"},"finish_reason":null}]}

661

635

662− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"21"},"finish_reason":null}]}

636+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

663

637

664− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" w"},"finish_reason":null}]}

638+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

665

639

666− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"22"},"finish_reason":null}]}

640+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

667

641

668− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" i"},"finish_reason":null}]}

642+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

669

643

670− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"23"},"finish_reason":null}]}

644+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Create"},"finish_reason":null}]}

671

645

672− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

646+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" test"},"finish_reason":null}]}

673

647

674− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"24"},"finish_reason":null}]}

648+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

675

649

676− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" h"},"finish_reason":null}]}

650+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

677

651

678− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"25"},"finish_reason":null}]}

652+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" hello"},"finish_reason":null}]}

679

653

680− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

654+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" bash"},"finish_reason":null}]}

681

655

682− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"26"},"finish_reason":null}]}

656+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

683

657

684− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" h"},"finish_reason":null}]}

658+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

685

659

686− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"27"},"finish_reason":null}]}

660+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" timestamp"},"finish_reason":null}]}

687

661

688− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

662+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

689

663

690− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"28"},"finish_reason":null}]}

664+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Characters"},"finish_reason":null}]}

691

665

692− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" l"},"finish_reason":null}]}

666+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

693

667

694− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"29"},"finish_reason":null}]}

668+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" C"},"finish_reason":null}]}

695

669

696− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" l"},"finish_reason":null}]}

670+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

697

671

698− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"30"},"finish_reason":null}]}

672+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" r"},"finish_reason":null}]}

699

673

700− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" o"},"finish_reason":null}]}

674+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"2"},"finish_reason":null}]}

701

675

702− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"31"},"finish_reason":null}]}

676+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

703

677

704− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

678+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"3"},"finish_reason":null}]}

705

679

706− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"32"},"finish_reason":null}]}

680+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

707

681

708− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" b"},"finish_reason":null}]}

682+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

709

683

710− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"33"},"finish_reason":null}]}

684+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

711

685

712− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

686+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"5"},"finish_reason":null}]}

713

687

714− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"34"},"finish_reason":null}]}

688+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

715

689

716− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" s"},"finish_reason":null}]}

690+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"6"},"finish_reason":null}]}

717

691

718− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"35"},"finish_reason":null}]}

692+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

719

693

720− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" h"},"finish_reason":null}]}

694+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"7"},"finish_reason":null}]}

721

695

722− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"36"},"finish_reason":null}]}

696+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

723

697

724− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

698+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"8"},"finish_reason":null}]}

725

699

726− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"37"},"finish_reason":null}]}

700+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

727

701

728− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" n"},"finish_reason":null}]}

702+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"9"},"finish_reason":null}]}

729

703

730− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"38"},"finish_reason":null}]}

704+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" s"},"finish_reason":null}]}

731

705

732− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" o"},"finish_reason":null}]}

706+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"10"},"finish_reason":null}]}

733

707

734− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"39"},"finish_reason":null}]}

708+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

735

709

736− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

710+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"11"},"finish_reason":null}]}

737

711

738− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"40"},"finish_reason":null}]}

712+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

739

713

740− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

714+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"12"},"finish_reason":null}]}

741

715

742− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"41"},"finish_reason":null}]}

716+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

743

717

744− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" i"},"finish_reason":null}]}

718+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"13"},"finish_reason":null}]}

745

719

746− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"42"},"finish_reason":null}]}

720+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" x"},"finish_reason":null}]}

747

721

748− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" m"},"finish_reason":null}]}

722+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"14"},"finish_reason":null}]}

749

723

750− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"43"},"finish_reason":null}]}

724+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

751

725

752− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

726+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"15"},"finish_reason":null}]}

753

727

754− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"44"},"finish_reason":null}]}

728+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

755

729

756− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" s"},"finish_reason":null}]}

730+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"16"},"finish_reason":null}]}

757

731

758− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"45"},"finish_reason":null}]}

732+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" w"},"finish_reason":null}]}

759

733

760− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

734+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"17"},"finish_reason":null}]}

761

735

762− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"46"},"finish_reason":null}]}

736+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" i"},"finish_reason":null}]}

763

737

764− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

738+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"18"},"finish_reason":null}]}

765

739

766− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"47"},"finish_reason":null}]}

740+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

767

741

768− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" m"},"finish_reason":null}]}

742+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"19"},"finish_reason":null}]}

769

743

770− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"48"},"finish_reason":null}]}

744+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" h"},"finish_reason":null}]}

771

745

772− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" p"},"finish_reason":null}]}

746+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"20"},"finish_reason":null}]}

773

747

774− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"49"},"finish_reason":null}]}

748+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

775

749

776− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

750+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"21"},"finish_reason":null}]}

777

751

778− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

752+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" h"},"finish_reason":null}]}

779

753

780− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

754+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"22"},"finish_reason":null}]}

781

755

782− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"49"},"finish_reason":null}]}

756+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

783

757

784− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

758+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"23"},"finish_reason":null}]}

785

759

786− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

760+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" l"},"finish_reason":null}]}

787

761

788− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"including"},"finish_reason":null}]}

762+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"24"},"finish_reason":null}]}

789

763

790− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" spaces"},"finish_reason":null}]}

764+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" l"},"finish_reason":null}]}

791

765

792− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")."},"finish_reason":null}]}

766+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"25"},"finish_reason":null}]}

793

767

794− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" No"},"finish_reason":null}]}

768+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" o"},"finish_reason":null}]}

795

769

796− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

770+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"26"},"finish_reason":null}]}

797

771

798− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

772+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

799

773

800− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

774+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"27"},"finish_reason":null}]}

801

775

802− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" colon"},"finish_reason":null}]}

776+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" b"},"finish_reason":null}]}

803

777

804− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

778+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"28"},"finish_reason":null}]}

805

779

806− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Good"},"finish_reason":null}]}

780+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

807

781

808− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".\n\n"},"finish_reason":null}]}

782+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"29"},"finish_reason":null}]}

809

783

810− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Return"},"finish_reason":null}]}

784+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" s"},"finish_reason":null}]}

811

785

812− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" it"},"finish_reason":null}]}

786+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"30"},"finish_reason":null}]}

813

787

814− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

788+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" h"},"finish_reason":null}]}

815

789

816− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"B"},"finish_reason":null}]}

790+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"31"},"finish_reason":null}]}

817

791

818− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"ash"},"finish_reason":null}]}

792+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ,"},"finish_reason":null}]}

819

793

820− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" create"},"finish_reason":null}]}

794+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"32"},"finish_reason":null}]}

821

795

822− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" test"},"finish_reason":null}]}

796+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

823

797

824− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":".txt"},"finish_reason":null}]}

798+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"33"},"finish_reason":null}]}

825

799

826− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" with"},"finish_reason":null}]}

800+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" n"},"finish_reason":null}]}

827

801

828− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

802+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"34"},"finish_reason":null}]}

829

803

830− data: {"id":"chatcmpl-13ec86e9670d0a79f264454b","object":"chat.completion.chunk","created":1779218868,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":208,"completion_tokens":413,"total_tokens":621,"completion_tokens_details":{"reasoning_tokens":296},"cost":{"usd":0.0003044,"hypercredits":0.0060880000000000005},"remaining":{"hypercredits":0}}}

804+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" o"},"finish_reason":null}]}

831

805

832− data: [DONE]

806+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"35"},"finish_reason":null}]}

833

807

834− headers:

808+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

835− Content-Type:

836− - text/event-stream

837− status: 200 OK

838− code: 200

839− duration: 507.202917ms

840−- id: 1

841− request:

842− proto: HTTP/1.1

843− proto_major: 1

844− proto_minor: 1

845− content_length: 34004

846− host: ""

847− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/bash_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use bash to create a file named test.txt with content 'hello bash'. do not print its timestamp\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

848− headers:

849− Accept:

850− - application/json

851− Content-Type:

852− - application/json

853− User-Agent:

854− - Charm-Crush/devel (https://charm.land/crush)

855− url: https://hyper.charm.land/v1/chat/completions

856− method: POST

857− response:

858− proto: HTTP/2.0

859− proto_major: 2

860− proto_minor: 0

861− content_length: -1

862− uncompressed: true

863− body: |+

864− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

865

809

866− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

810+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"36"},"finish_reason":null}]}

867

811

868− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

812+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

869

813

870− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" wants me"},"finish_reason":null}]}

814+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"37"},"finish_reason":null}]}

871

815

872− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" to create a file"},"finish_reason":null}]}

816+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" i"},"finish_reason":null}]}

873

817

874− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" named"},"finish_reason":null}]}

818+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"38"},"finish_reason":null}]}

875

819

876− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" test"},"finish_reason":null}]}

820+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" m"},"finish_reason":null}]}

877

821

878− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":".txt with"},"finish_reason":null}]}

822+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"39"},"finish_reason":null}]}

879

823

880− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" content '"},"finish_reason":null}]}

824+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

881

825

882− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"hello bash"},"finish_reason":null}]}

826+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"40"},"finish_reason":null}]}

883

827

884− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"' using"},"finish_reason":null}]}

828+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" s"},"finish_reason":null}]}

885

829

886− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

830+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"41"},"finish_reason":null}]}

887

831

888− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" bash tool"},"finish_reason":null}]}

832+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

889

833

890− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":", and not"},"finish_reason":null}]}

834+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"42"},"finish_reason":null}]}

891

835

892− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" print its"},"finish_reason":null}]}

836+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

893

837

894− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" timestamp."},"finish_reason":null}]}

838+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"43"},"finish_reason":null}]}

895

839

896− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_95f948f5d1574cbcb5abcdec","type":"function","index":0,"function":{"name":"bash","arguments":""}}]},"finish_reason":null}]}

840+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" m"},"finish_reason":null}]}

897

841

898− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

842+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"44"},"finish_reason":null}]}

899

843

900− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

844+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" p"},"finish_reason":null}]}

901

845

902− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"command\": "}}]},"finish_reason":null}]}

846+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"45"},"finish_reason":null}]}

903

847

904− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"printf"}}]},"finish_reason":null}]}

848+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

905

849

906− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" 'hello"}}]},"finish_reason":null}]}

850+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

907

851

908− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" bash'"}}]},"finish_reason":null}]}

852+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

909

853

910− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" \u003e /"}}]},"finish_reason":null}]}

854+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"45"},"finish_reason":null}]}

911

855

912− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"tmp"}}]},"finish_reason":null}]}

856+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

913

857

914− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/c"}}]},"finish_reason":null}]}

858+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

915

859

916− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"rush"}}]},"finish_reason":null}]}

860+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Good"},"finish_reason":null}]}

917

861

918− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-test/Test"}}]},"finish_reason":null}]}

862+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".\n\n"},"finish_reason":null}]}

919

863

920− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Coder"}}]},"finish_reason":null}]}

864+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Return"},"finish_reason":null}]}

921

865

922− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Agent"}}]},"finish_reason":null}]}

866+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" that"},"finish_reason":null}]}

923

867

924− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/glm"}}]},"finish_reason":null}]}

868+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

925

869

926− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-5"}}]},"finish_reason":null}]}

870+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

927

871

928− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".1"}}]},"finish_reason":null}]}

872+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Create"},"finish_reason":null}]}

929

873

930− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/bash_tool"}}]},"finish_reason":null}]}

874+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" test"},"finish_reason":null}]}

931

875

932− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/test.txt\""}}]},"finish_reason":null}]}

876+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":".txt"},"finish_reason":null}]}

933

877

934− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":", "}}]},"finish_reason":null}]}

878+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" with"},"finish_reason":null}]}

935

879

936− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"description\": "}}]},"finish_reason":null}]}

880+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" hello"},"finish_reason":null}]}

937

881

938− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"Create test"}}]},"finish_reason":null}]}

882+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" bash"},"finish_reason":null}]}

939

883

940− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".txt with content"}}]},"finish_reason":null}]}

884+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":","},"finish_reason":null}]}

941

885

942− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\""}}]},"finish_reason":null}]}

886+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" no"},"finish_reason":null}]}

943

887

944− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"}"}}]},"finish_reason":null}]}

888+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" timestamp"},"finish_reason":null}]}

945

889

946− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

890+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

947

891

948− data: {"id":"chatcmpl-512f4719c31d4e4070f0c744","object":"chat.completion.chunk","created":1779218869,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":720,"completion_tokens":72,"total_tokens":7896,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":28},"cost":{"usd":0.00152,"hypercredits":0.0304},"remaining":{"hypercredits":0}}}

892+ data: {"id":"chatcmpl-3f896991098deae412ae2685","object":"chat.completion.chunk","created":1780069679,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":224,"completion_tokens":374,"total_tokens":598,"completion_tokens_details":{"reasoning_tokens":356},"cost":{"usd":0.00048562,"hypercredits":0.0097124},"remaining":{"hypercredits":0}}}

949

893

950 data: [DONE]

894 data: [DONE]

951

895

@@ -954,15 +898,15 @@ interactions:

954 - text/event-stream

898 - text/event-stream

955 status: 200 OK

899 status: 200 OK

956 code: 200

900 code: 200

957− duration: 1.339518333s

901+ duration: 439.758791ms

958- id: 2

902- id: 2

959 request:

903 request:

960 proto: HTTP/1.1

904 proto: HTTP/1.1

961 proto_major: 1

905 proto_major: 1

962 proto_minor: 1

906 proto_minor: 1

963− content_length: 34521

907+ content_length: 34451

964 host: ""

908 host: ""

965− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/bash_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use bash to create a file named test.txt with content 'hello bash'. do not print its timestamp\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_95f948f5d1574cbcb5abcdec\",\"function\":{\"arguments\":\"{\\\"command\\\": \\\"printf 'hello bash' > /tmp/crush-test/TestCoderAgent/glm-5.1/bash_tool/test.txt\\\", \\\"description\\\": \\\"Create test.txt with content\\\"}\",\"name\":\"bash\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"The user wants me to create a file named test.txt with content 'hello bash' using the bash tool, and not print its timestamp.\"},{\"content\":\"no output\",\"tool_call_id\":\"call_95f948f5d1574cbcb5abcdec\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

909+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/bash_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use bash to create a file named test.txt with content 'hello bash'. do not print its timestamp\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_maHr9QegPpRMRiFEH9e4bSTo\",\"function\":{\"arguments\":\"{\\\"command\\\": \\\"echo 'hello bash' > test.txt\\\", \\\"description\\\": \\\"Create test.txt with content\\\"}\",\"name\":\"bash\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"The user wants me to create a file named test.txt with content 'hello bash' using the bash tool, and not print its timestamp.\"},{\"content\":\"no output\",\"tool_call_id\":\"call_maHr9QegPpRMRiFEH9e4bSTo\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

966 headers:

910 headers:

967 Accept:

911 Accept:

968 - application/json

912 - application/json

@@ -979,25 +923,15 @@ interactions:

979 content_length: -1

923 content_length: -1

980 uncompressed: true

924 uncompressed: true

981 body: |+

925 body: |+

982− data: {"id":"chatcmpl-d7168cb873f32f392e749ff0","object":"chat.completion.chunk","created":1779218873,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

926+ data: {"id":"chatcmpl-4ab91f0c6c3e7671bcf2a109","object":"chat.completion.chunk","created":1780069682,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

983−

984− data: {"id":"chatcmpl-d7168cb873f32f392e749ff0","object":"chat.completion.chunk","created":1779218873,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Created"},"finish_reason":null}]}

985−

986− data: {"id":"chatcmpl-d7168cb873f32f392e749ff0","object":"chat.completion.chunk","created":1779218873,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

987−

988− data: {"id":"chatcmpl-d7168cb873f32f392e749ff0","object":"chat.completion.chunk","created":1779218873,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"test.txt"},"finish_reason":null}]}

989−

990− data: {"id":"chatcmpl-d7168cb873f32f392e749ff0","object":"chat.completion.chunk","created":1779218873,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"` with"},"finish_reason":null}]}

991−

992− data: {"id":"chatcmpl-d7168cb873f32f392e749ff0","object":"chat.completion.chunk","created":1779218873,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" content `"},"finish_reason":null}]}

993

927

994− data: {"id":"chatcmpl-d7168cb873f32f392e749ff0","object":"chat.completion.chunk","created":1779218873,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"hello bash"},"finish_reason":null}]}

928+ data: {"id":"chatcmpl-4ab91f0c6c3e7671bcf2a109","object":"chat.completion.chunk","created":1780069682,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Done"},"finish_reason":null}]}

995

929

996− data: {"id":"chatcmpl-d7168cb873f32f392e749ff0","object":"chat.completion.chunk","created":1779218873,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`."},"finish_reason":null}]}

930+ data: {"id":"chatcmpl-4ab91f0c6c3e7671bcf2a109","object":"chat.completion.chunk","created":1780069682,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"."},"finish_reason":null}]}

997

931

998− data: {"id":"chatcmpl-d7168cb873f32f392e749ff0","object":"chat.completion.chunk","created":1779218873,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

932+ data: {"id":"chatcmpl-4ab91f0c6c3e7671bcf2a109","object":"chat.completion.chunk","created":1780069682,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

999

933

1000− data: {"id":"chatcmpl-d7168cb873f32f392e749ff0","object":"chat.completion.chunk","created":1779218873,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":66,"completion_tokens":13,"total_tokens":7887,"prompt_tokens_details":{"cached_tokens":7808},"completion_tokens_details":{"reasoning_tokens":12},"cost":{"usd":0.000209,"hypercredits":0.00418},"remaining":{"hypercredits":0}}}

934+ data: {"id":"chatcmpl-4ab91f0c6c3e7671bcf2a109","object":"chat.completion.chunk","created":1780069682,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":45,"completion_tokens":4,"total_tokens":7857,"prompt_tokens_details":{"cached_tokens":7808},"cost":{"usd":0.0001,"hypercredits":0.002},"remaining":{"hypercredits":0}}}

1001

935

1002 data: [DONE]

936 data: [DONE]

1003

937

@@ -1006,4 +940,4 @@ interactions:

1006 - text/event-stream

940 - text/event-stream

1007 status: 200 OK

941 status: 200 OK

1008 code: 200

942 code: 200

1009− duration: 1.985415708s

943+ duration: 2.925894959s

internal/agent/testdata/TestCoderAgent/glm-5.1/download_tool.yaml+599−173

@@ -2,6 +2,152 @@

2version: 2

3interactions:

4- id: 0

5+ request:

6+ proto: HTTP/1.1

7+ proto_major: 1

8+ proto_minor: 1

9+ content_length: 34010

10+ host: ""

11+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/download_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"download the file from https://example-files.online-convert.com/document/txt/example.txt and save it as example.txt\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

12+ headers:

13+ Accept:

14+ - application/json

15+ Content-Type:

16+ - application/json

17+ User-Agent:

18+ - Charm-Crush/devel (https://charm.land/crush)

19+ url: https://hyper.charm.land/v1/chat/completions

20+ method: POST

21+ response:

22+ proto: HTTP/2.0

23+ proto_major: 2

24+ proto_minor: 0

25+ content_length: -1

26+ uncompressed: true

27+ body: |+

28+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

29+

30+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

31+

32+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

33+

34+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" wants"},"finish_reason":null}]}

35+

36+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" me"},"finish_reason":null}]}

37+

38+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

39+

40+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" download"},"finish_reason":null}]}

41+

42+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

43+

44+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" file"},"finish_reason":null}]}

45+

46+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" from"},"finish_reason":null}]}

47+

48+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

49+

50+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" URL"},"finish_reason":null}]}

51+

52+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

53+

54+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" save"},"finish_reason":null}]}

55+

56+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" it"},"finish_reason":null}]}

57+

58+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" as"},"finish_reason":null}]}

59+

60+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" example"},"finish_reason":null}]}

61+

62+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

63+

64+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

65+

66+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_LiiFGSBspKNdPTWMEMOLRvLr","type":"function","index":0,"function":{"name":"download","arguments":""}}]},"finish_reason":null}]}

67+

68+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

69+

70+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

71+

72+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"file_path\": "}}]},"finish_reason":null}]}

73+

74+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"/tmp"}}]},"finish_reason":null}]}

75+

76+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/c"}}]},"finish_reason":null}]}

77+

78+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"rush"}}]},"finish_reason":null}]}

79+

80+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-test"}}]},"finish_reason":null}]}

81+

82+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/Test"}}]},"finish_reason":null}]}

83+

84+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Coder"}}]},"finish_reason":null}]}

85+

86+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Agent"}}]},"finish_reason":null}]}

87+

88+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/g"}}]},"finish_reason":null}]}

89+

90+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"lm"}}]},"finish_reason":null}]}

91+

92+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-"}}]},"finish_reason":null}]}

93+

94+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"5"}}]},"finish_reason":null}]}

95+

96+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"."}}]},"finish_reason":null}]}

97+

98+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"1"}}]},"finish_reason":null}]}

99+

100+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/download"}}]},"finish_reason":null}]}

101+

102+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_tool"}}]},"finish_reason":null}]}

103+

104+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/example"}}]},"finish_reason":null}]}

105+

106+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".txt"}}]},"finish_reason":null}]}

107+

108+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\", \"url\": "}}]},"finish_reason":null}]}

109+

110+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"https"}}]},"finish_reason":null}]}

111+

112+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"://"}}]},"finish_reason":null}]}

113+

114+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"example"}}]},"finish_reason":null}]}

115+

116+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-files"}}]},"finish_reason":null}]}

117+

118+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".online"}}]},"finish_reason":null}]}

119+

120+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-"}}]},"finish_reason":null}]}

121+

122+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"convert"}}]},"finish_reason":null}]}

123+

124+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".com"}}]},"finish_reason":null}]}

125+

126+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/document"}}]},"finish_reason":null}]}

127+

128+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/t"}}]},"finish_reason":null}]}

129+

130+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"xt"}}]},"finish_reason":null}]}

131+

132+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/example"}}]},"finish_reason":null}]}

133+

134+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".txt"}}]},"finish_reason":null}]}

135+

136+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"}"}}]},"finish_reason":null}]}

137+

138+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

139+

140+ data: {"id":"chatcmpl-b24616267883946383972241","object":"chat.completion.chunk","created":1780069683,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":718,"completion_tokens":64,"total_tokens":7886,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":19},"cost":{"usd":0.00128918,"hypercredits":0.0257836},"remaining":{"hypercredits":0}}}

141+

142+ data: [DONE]

143+

144+ headers:

145+ Content-Type:

146+ - text/event-stream

147+ status: 200 OK

148+ code: 200

149+ duration: 638.4435ms

150+- id: 1

5 request:

151 request:

6 proto: HTTP/1.1

152 proto: HTTP/1.1

7 proto_major: 1

153 proto_major: 1

@@ -25,291 +171,567 @@ interactions:

25 content_length: -1

171 content_length: -1

26 uncompressed: true

172 uncompressed: true

27 body: |+

173 body: |+

28− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

174+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

175

30− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

176+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"User"},"finish_reason":null}]}

177

32− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

178+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" wants"},"finish_reason":null}]}

179

34− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" asks"},"finish_reason":null}]}

180+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

181

36− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

182+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

183

38− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

184+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

185

40− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Generate"},"finish_reason":null}]}

186+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" summar"},"finish_reason":null}]}

187

42− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

188+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"izing"},"finish_reason":null}]}

189

44− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

190+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

191

46− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

192+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

193

48− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

194+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"download"},"finish_reason":null}]}

195

50− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

196+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

197

52− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" following"},"finish_reason":null}]}

198+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" file"},"finish_reason":null}]}

199

54− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

200+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" from"},"finish_reason":null}]}

201

56− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":\n\n"},"finish_reason":null}]}

202+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" https"},"finish_reason":null}]}

203

58− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"download"},"finish_reason":null}]}

204+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"://"},"finish_reason":null}]}

205

60− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

206+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"..."},"finish_reason":null}]}

207

62− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" file"},"finish_reason":null}]}

208+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

209

64− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" from"},"finish_reason":null}]}

210+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" save"},"finish_reason":null}]}

211

66− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" https"},"finish_reason":null}]}

212+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" it"},"finish_reason":null}]}

213

68− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"://"},"finish_reason":null}]}

214+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" as"},"finish_reason":null}]}

215

70− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"example"},"finish_reason":null}]}

216+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" example"},"finish_reason":null}]}

217

72− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"-files"},"finish_reason":null}]}

218+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

219

74− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".online"},"finish_reason":null}]}

220+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

221

76− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"-con"},"finish_reason":null}]}

222+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Must"},"finish_reason":null}]}

223

78− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"vert"},"finish_reason":null}]}

224+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

225

80− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".com"},"finish_reason":null}]}

226+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

227

82− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"/document"},"finish_reason":null}]}

228+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" same"},"finish_reason":null}]}

229

84− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"/t"},"finish_reason":null}]}

230+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" language"},"finish_reason":null}]}

231

86− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"xt"},"finish_reason":null}]}

232+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

233

88− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"/example"},"finish_reason":null}]}

234+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"English"},"finish_reason":null}]}

235

90− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

236+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")."},"finish_reason":null}]}

237

92− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

238+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \u003c="},"finish_reason":null}]}

239

94− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" save"},"finish_reason":null}]}

240+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

241

96− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" it"},"finish_reason":null}]}

242+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

243

98− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" as"},"finish_reason":null}]}

244+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

245

100− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" example"},"finish_reason":null}]}

246+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" One"},"finish_reason":null}]}

101

247

102− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

248+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

103

249

104− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

250+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

105

251

106− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Need"},"finish_reason":null}]}

252+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

107

253

108− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

254+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

109

255

110− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" short"},"finish_reason":null}]}

256+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" or"},"finish_reason":null}]}

111

257

112− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

258+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" col"},"finish_reason":null}]}

113

259

114− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

260+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ons"},"finish_reason":null}]}

115

261

116− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" same"},"finish_reason":null}]}

262+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".\n\n"},"finish_reason":null}]}

117

263

118− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" language"},"finish_reason":null}]}

264+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Possible"},"finish_reason":null}]}

119

265

120− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

266+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

121

267

122− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"English"},"finish_reason":null}]}

268+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

123

269

124− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")."},"finish_reason":null}]}

270+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

125

271

126− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Under"},"finish_reason":null}]}

272+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Download"},"finish_reason":null}]}

127

273

128− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

274+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" example"},"finish_reason":null}]}

129

275

130− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

276+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

131

277

132− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

278+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" from"},"finish_reason":null}]}

133

279

134− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

280+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" URL"},"finish_reason":null}]}

135

281

136− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Something"},"finish_reason":null}]}

282+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

137

283

138− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" like"},"finish_reason":null}]}

284+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" save"},"finish_reason":null}]}

139

285

140− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

286+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" locally"},"finish_reason":null}]}

141

287

142− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Download"},"finish_reason":null}]}

288+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

143

289

144− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

290+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

145

291

146− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Save"},"finish_reason":null}]}

292+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

147

293

148− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" example"},"finish_reason":null}]}

294+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

149

295

150− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

296+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

151

297

152− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

298+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Download"},"finish_reason":null}]}

153

299

154− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

300+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" example"},"finish_reason":null}]}

155

301

156− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

302+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

157

303

158− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

304+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" from"},"finish_reason":null}]}

159

305

160− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

306+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" URL"},"finish_reason":null}]}

161

307

162− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Download"},"finish_reason":null}]}

308+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

163

309

164− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

310+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" save"},"finish_reason":null}]}

165

311

166− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Save"},"finish_reason":null}]}

312+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" locally"},"finish_reason":null}]}

167

313

168− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" example"},"finish_reason":null}]}

314+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

169

315

170− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

316+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ="},"finish_reason":null}]}

171

317

172− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

318+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" let's"},"finish_reason":null}]}

173

319

174− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ="},"finish_reason":null}]}

320+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

175

321

176− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

322+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

177

323

178− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"29"},"finish_reason":null}]}

324+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Download"},"finish_reason":null}]}

179

325

180− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

326+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

181

327

182− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Good"},"finish_reason":null}]}

328+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"8"},"finish_reason":null}]}

183

329

184− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".\n\n"},"finish_reason":null}]}

330+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

185

331

186− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Return"},"finish_reason":null}]}

332+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" +"},"finish_reason":null}]}

187

333

188− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" only"},"finish_reason":null}]}

334+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

189

335

190− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

336+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

191

337

192− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

338+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

193

339

194− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Download"},"finish_reason":null}]}

340+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" =\u003e"},"finish_reason":null}]}

195

341

196− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" and"},"finish_reason":null}]}

342+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"9"},"finish_reason":null}]}

197

343

198− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" Save"},"finish_reason":null}]}

344+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

199

345

200− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" example"},"finish_reason":null}]}

346+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" example"},"finish_reason":null}]}

201

347

202− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":".txt"},"finish_reason":null}]}

348+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

203

349

204− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

350+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

205

351

206− data: {"id":"chatcmpl-efdcae4864e03517d94febb1","object":"chat.completion.chunk","created":1779218874,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":165,"completion_tokens":5,"total_tokens":170,"completion_tokens_details":{"reasoning_tokens":82},"cost":{"usd":0.0001,"hypercredits":0.002},"remaining":{"hypercredits":0}}}

352+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"11"},"finish_reason":null}]}

207

353

208− data: [DONE]

354+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

209

355

210− headers:

356+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" =\u003e"},"finish_reason":null}]}

211− Content-Type:

357+

212− - text/event-stream

358+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"20"},"finish_reason":null}]}

213− status: 200 OK

359+

214− code: 200

360+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

215− duration: 709.584208ms

361+

216−- id: 1

362+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

217− request:

363+

218− proto: HTTP/1.1

364+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

219− proto_major: 1

365+

220− proto_minor: 1

366+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

221− content_length: 34029

367+

222− host: ""

368+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" =\u003e"},"finish_reason":null}]}

223− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/download_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"download the file from https://example-files.online-convert.com/document/txt/example.txt and save it as example.txt\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

369+

224− headers:

370+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"21"},"finish_reason":null}]}

225− Accept:

371+

226− - application/json

372+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

227− Content-Type:

373+

228− - application/json

374+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" from"},"finish_reason":null}]}

229− User-Agent:

375+

230− - Charm-Crush/devel (https://charm.land/crush)

376+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

231− url: https://hyper.charm.land/v1/chat/completions

377+

232− method: POST

378+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

233− response:

379+

234− proto: HTTP/2.0

380+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")=\u003e"},"finish_reason":null}]}

235− proto_major: 2

381+

236− proto_minor: 0

382+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"25"},"finish_reason":null}]}

237− content_length: -1

383+

238− uncompressed: true

384+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

239− body: |+

385+

240− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

386+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

387+

388+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

389+

390+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

391+

392+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"=\u003e"},"finish_reason":null}]}

393+

394+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"26"},"finish_reason":null}]}

395+

396+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

397+

398+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" URL"},"finish_reason":null}]}

241

399

242− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

400+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

243

401

244− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

402+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"3"},"finish_reason":null}]}

245

403

246− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" wants me"},"finish_reason":null}]}

404+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")=\u003e"},"finish_reason":null}]}

247

405

248− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" to download a"},"finish_reason":null}]}

406+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"29"},"finish_reason":null}]}

249

407

250− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" file from a"},"finish_reason":null}]}

408+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

251

409

252− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" URL and"},"finish_reason":null}]}

410+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

253

411

254− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" save it"},"finish_reason":null}]}

412+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

255

413

256− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" as example"},"finish_reason":null}]}

414+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

257

415

258− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

416+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"=\u003e"},"finish_reason":null}]}

259

417

260− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

418+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"30"},"finish_reason":null}]}

261

419

262− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_2b8e8e5bf00b4ca1b7a0b856","type":"function","index":0,"function":{"name":"download","arguments":""}}]},"finish_reason":null}]}

420+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

263

421

264− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

422+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

265

423

266− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

424+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

267

425

268− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"file_path\": \"/tmp"}}]},"finish_reason":null}]}

426+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"3"},"finish_reason":null}]}

269

427

270− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/c"}}]},"finish_reason":null}]}

428+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")=\u003e"},"finish_reason":null}]}

271

429

272− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"rush"}}]},"finish_reason":null}]}

430+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"33"},"finish_reason":null}]}

273

431

274− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-test"}}]},"finish_reason":null}]}

432+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

275

433

276− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/Test"}}]},"finish_reason":null}]}

434+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

277

435

278− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Coder"}}]},"finish_reason":null}]}

436+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

279

437

280− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Agent"}}]},"finish_reason":null}]}

438+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

281

439

282− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/glm"}}]},"finish_reason":null}]}

440+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"=\u003e"},"finish_reason":null}]}

283

441

284− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-5"}}]},"finish_reason":null}]}

442+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"34"},"finish_reason":null}]}

285

443

286− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".1"}}]},"finish_reason":null}]}

444+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

287

445

288− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/download_tool"}}]},"finish_reason":null}]}

446+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" save"},"finish_reason":null}]}

289

447

290− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/example.txt\""}}]},"finish_reason":null}]}

448+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

291

449

292− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":", "}}]},"finish_reason":null}]}

450+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

293

451

294− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"url\": \"https"}}]},"finish_reason":null}]}

452+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")=\u003e"},"finish_reason":null}]}

295

453

296− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"://example-files"}}]},"finish_reason":null}]}

454+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"38"},"finish_reason":null}]}

297

455

298− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".online-"}}]},"finish_reason":null}]}

456+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

299

457

300− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"convert.com/document"}}]},"finish_reason":null}]}

458+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

301

459

302− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/t"}}]},"finish_reason":null}]}

460+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

303

461

304− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"xt/example"}}]},"finish_reason":null}]}

462+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

305

463

306− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".txt\""}}]},"finish_reason":null}]}

464+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"=\u003e"},"finish_reason":null}]}

307

465

308− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"}"}}]},"finish_reason":null}]}

466+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"39"},"finish_reason":null}]}

309

467

310− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

468+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

311

469

312− data: {"id":"chatcmpl-24223fc1befa78013925ba25","object":"chat.completion.chunk","created":1779218876,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":723,"completion_tokens":64,"total_tokens":7891,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":18},"cost":{"usd":0.0014452999999999998,"hypercredits":0.028905999999999994},"remaining":{"hypercredits":0}}}

470+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" locally"},"finish_reason":null}]}

471+

472+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

473+

474+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"7"},"finish_reason":null}]}

475+

476+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")=\u003e"},"finish_reason":null}]}

477+

478+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"46"},"finish_reason":null}]}

479+

480+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

481+

482+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" So"},"finish_reason":null}]}

483+

484+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

485+

486+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"46"},"finish_reason":null}]}

487+

488+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

489+

490+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

491+

492+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" okay"},"finish_reason":null}]}

493+

494+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".\n\n"},"finish_reason":null}]}

495+

496+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Alternatively"},"finish_reason":null}]}

497+

498+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

499+

500+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

501+

502+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Download"},"finish_reason":null}]}

503+

504+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" file"},"finish_reason":null}]}

505+

506+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" from"},"finish_reason":null}]}

507+

508+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" URL"},"finish_reason":null}]}

509+

510+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

511+

512+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" save"},"finish_reason":null}]}

513+

514+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" as"},"finish_reason":null}]}

515+

516+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" example"},"finish_reason":null}]}

517+

518+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

519+

520+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

521+

522+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

523+

524+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

525+

526+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Download"},"finish_reason":null}]}

527+

528+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

529+

530+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"8"},"finish_reason":null}]}

531+

532+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")+"},"finish_reason":null}]}

533+

534+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

535+

536+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

537+

538+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

539+

540+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"9"},"finish_reason":null}]}

541+

542+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

543+

544+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" file"},"finish_reason":null}]}

545+

546+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

547+

548+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

549+

550+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

551+

552+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"13"},"finish_reason":null}]}

553+

554+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

555+

556+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

557+

558+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

559+

560+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

561+

562+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"14"},"finish_reason":null}]}

563+

564+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

565+

566+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" from"},"finish_reason":null}]}

567+

568+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

569+

570+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

571+

572+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

573+

574+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"18"},"finish_reason":null}]}

575+

576+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

577+

578+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

579+

580+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

581+

582+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

583+

584+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"19"},"finish_reason":null}]}

585+

586+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

587+

588+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" URL"},"finish_reason":null}]}

589+

590+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

591+

592+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"3"},"finish_reason":null}]}

593+

594+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

595+

596+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"22"},"finish_reason":null}]}

597+

598+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

599+

600+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

601+

602+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

603+

604+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

605+

606+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"23"},"finish_reason":null}]}

607+

608+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

609+

610+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

611+

612+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

613+

614+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"3"},"finish_reason":null}]}

615+

616+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

617+

618+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"26"},"finish_reason":null}]}

619+

620+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

621+

622+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

623+

624+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

625+

626+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

627+

628+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"27"},"finish_reason":null}]}

629+

630+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

631+

632+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" save"},"finish_reason":null}]}

633+

634+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

635+

636+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

637+

638+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

639+

640+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"31"},"finish_reason":null}]}

641+

642+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

643+

644+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

645+

646+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

647+

648+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

649+

650+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"32"},"finish_reason":null}]}

651+

652+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

653+

654+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" as"},"finish_reason":null}]}

655+

656+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

657+

658+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"2"},"finish_reason":null}]}

659+

660+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

661+

662+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"34"},"finish_reason":null}]}

663+

664+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

665+

666+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

667+

668+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

669+

670+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

671+

672+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"35"},"finish_reason":null}]}

673+

674+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

675+

676+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" example"},"finish_reason":null}]}

677+

678+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".txt"},"finish_reason":null}]}

679+

680+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

681+

682+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"11"},"finish_reason":null}]}

683+

684+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

685+

686+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"46"},"finish_reason":null}]}

687+

688+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

689+

690+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" So"},"finish_reason":null}]}

691+

692+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" also"},"finish_reason":null}]}

693+

694+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

695+

696+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"46"},"finish_reason":null}]}

697+

698+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".\n\n"},"finish_reason":null}]}

699+

700+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"We"},"finish_reason":null}]}

701+

702+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" need"},"finish_reason":null}]}

703+

704+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

705+

706+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

707+

708+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Choose"},"finish_reason":null}]}

709+

710+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" one"},"finish_reason":null}]}

711+

712+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

713+

714+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Download"},"finish_reason":null}]}

715+

716+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" file"},"finish_reason":null}]}

717+

718+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" from"},"finish_reason":null}]}

719+

720+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" URL"},"finish_reason":null}]}

721+

722+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" and"},"finish_reason":null}]}

723+

724+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" save"},"finish_reason":null}]}

725+

726+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" as"},"finish_reason":null}]}

727+

728+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" example"},"finish_reason":null}]}

729+

730+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":".txt"},"finish_reason":null}]}

731+

732+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

733+

734+ data: {"id":"chatcmpl-3f0f8f2b9b05cd8f7a280911","object":"chat.completion.chunk","created":1780069683,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":227,"completion_tokens":288,"total_tokens":515,"completion_tokens_details":{"reasoning_tokens":270},"cost":{"usd":0.00038124,"hypercredits":0.007624799999999999},"remaining":{"hypercredits":0}}}

313

735

314 data: [DONE]

736 data: [DONE]

315

737

@@ -318,7 +740,7 @@ interactions:

318 - text/event-stream

740 - text/event-stream

319 status: 200 OK

741 status: 200 OK

320 code: 200

742 code: 200

321− duration: 2.913161542s

743+ duration: 837.561291ms

322- id: 2

744- id: 2

323 request:

745 request:

324 proto: HTTP/1.1

746 proto: HTTP/1.1

@@ -364,15 +786,15 @@ interactions:

364 - text/plain; charset=UTF-8

786 - text/plain; charset=UTF-8

365 status: 200 OK

787 status: 200 OK

366 code: 200

788 code: 200

367− duration: 450.142541ms

789+ duration: 610.98125ms

368- id: 3

790- id: 3

369 request:

791 request:

370 proto: HTTP/1.1

792 proto: HTTP/1.1

371 proto_major: 1

793 proto_major: 1

372 proto_minor: 1

794 proto_minor: 1

373− content_length: 34598

795+ content_length: 34579

374 host: ""

796 host: ""

375− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/download_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"download the file from https://example-files.online-convert.com/document/txt/example.txt and save it as example.txt\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_2b8e8e5bf00b4ca1b7a0b856\",\"function\":{\"arguments\":\"{\\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/download_tool/example.txt\\\", \\\"url\\\": \\\"https://example-files.online-convert.com/document/txt/example.txt\\\"}\",\"name\":\"download\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"The user wants me to download a file from a URL and save it as example.txt.\"},{\"content\":\"Successfully downloaded 2574 bytes to example.txt (Content-Type: text/plain; charset=UTF-8)\",\"tool_call_id\":\"call_2b8e8e5bf00b4ca1b7a0b856\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

797+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/download_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"download the file from https://example-files.online-convert.com/document/txt/example.txt and save it as example.txt\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_LiiFGSBspKNdPTWMEMOLRvLr\",\"function\":{\"arguments\":\"{\\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/download_tool/example.txt\\\", \\\"url\\\": \\\"https://example-files.online-convert.com/document/txt/example.txt\\\"}\",\"name\":\"download\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"The user wants me to download a file from a URL and save it as example.txt.\"},{\"content\":\"Successfully downloaded 2574 bytes to example.txt (Content-Type: text/plain; charset=UTF-8)\",\"tool_call_id\":\"call_LiiFGSBspKNdPTWMEMOLRvLr\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

376 headers:

798 headers:

377 Accept:

799 Accept:

378 - application/json

800 - application/json

@@ -389,25 +811,29 @@ interactions:

389 content_length: -1

811 content_length: -1

390 uncompressed: true

812 uncompressed: true

391 body: |+

813 body: |+

392− data: {"id":"chatcmpl-e0b664475e72b448ffb799fb","object":"chat.completion.chunk","created":1779218879,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

814+ data: {"id":"chatcmpl-a709ede8c6d38d00517c62ad","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

815+

816+ data: {"id":"chatcmpl-a709ede8c6d38d00517c62ad","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Download"},"finish_reason":null}]}

817+

818+ data: {"id":"chatcmpl-a709ede8c6d38d00517c62ad","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"ed"},"finish_reason":null}]}

393

819

394− data: {"id":"chatcmpl-e0b664475e72b448ffb799fb","object":"chat.completion.chunk","created":1779218879,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Download"},"finish_reason":null}]}

820+ data: {"id":"chatcmpl-a709ede8c6d38d00517c62ad","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" and"},"finish_reason":null}]}

395

821

396− data: {"id":"chatcmpl-e0b664475e72b448ffb799fb","object":"chat.completion.chunk","created":1779218879,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"ed and"},"finish_reason":null}]}

822+ data: {"id":"chatcmpl-a709ede8c6d38d00517c62ad","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" saved"},"finish_reason":null}]}

397

823

398− data: {"id":"chatcmpl-e0b664475e72b448ffb799fb","object":"chat.completion.chunk","created":1779218879,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" saved as"},"finish_reason":null}]}

824+ data: {"id":"chatcmpl-a709ede8c6d38d00517c62ad","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" as"},"finish_reason":null}]}

399

825

400− data: {"id":"chatcmpl-e0b664475e72b448ffb799fb","object":"chat.completion.chunk","created":1779218879,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `example"},"finish_reason":null}]}

826+ data: {"id":"chatcmpl-a709ede8c6d38d00517c62ad","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

401

827

402− data: {"id":"chatcmpl-e0b664475e72b448ffb799fb","object":"chat.completion.chunk","created":1779218879,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".txt` (25"},"finish_reason":null}]}

828+ data: {"id":"chatcmpl-a709ede8c6d38d00517c62ad","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"example"},"finish_reason":null}]}

403

829

404− data: {"id":"chatcmpl-e0b664475e72b448ffb799fb","object":"chat.completion.chunk","created":1779218879,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"74"},"finish_reason":null}]}

830+ data: {"id":"chatcmpl-a709ede8c6d38d00517c62ad","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".txt"},"finish_reason":null}]}

405

831

406− data: {"id":"chatcmpl-e0b664475e72b448ffb799fb","object":"chat.completion.chunk","created":1779218879,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" bytes)."},"finish_reason":null}]}

832+ data: {"id":"chatcmpl-a709ede8c6d38d00517c62ad","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`."},"finish_reason":null}]}

407

833

408− data: {"id":"chatcmpl-e0b664475e72b448ffb799fb","object":"chat.completion.chunk","created":1779218879,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

834+ data: {"id":"chatcmpl-a709ede8c6d38d00517c62ad","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

409

835

410− data: {"id":"chatcmpl-e0b664475e72b448ffb799fb","object":"chat.completion.chunk","created":1779218879,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":91,"completion_tokens":17,"total_tokens":7916,"prompt_tokens_details":{"cached_tokens":7808},"completion_tokens_details":{"reasoning_tokens":16},"cost":{"usd":0.00028170000000000007,"hypercredits":0.005634000000000001},"remaining":{"hypercredits":0}}}

836+ data: {"id":"chatcmpl-a709ede8c6d38d00517c62ad","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":86,"completion_tokens":11,"total_tokens":7905,"prompt_tokens_details":{"cached_tokens":7808},"cost":{"usd":0.00015886,"hypercredits":0.0031772},"remaining":{"hypercredits":0}}}

411

837

412 data: [DONE]

838 data: [DONE]

413

839

@@ -416,4 +842,4 @@ interactions:

416 - text/event-stream

842 - text/event-stream

417 status: 200 OK

843 status: 200 OK

418 code: 200

844 code: 200

419− duration: 1.816622208s

845+ duration: 562.899125ms

internal/agent/testdata/TestCoderAgent/glm-5.1/fetch_tool.yaml+243−305

@@ -25,415 +25,311 @@ interactions:

25 content_length: -1

26 uncompressed: true

27 body: |+

28− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

28+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

30− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

30+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

32− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

32+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

34− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" wants"},"finish_reason":null}]}

34+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" wants"},"finish_reason":null}]}

36− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

36+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

38− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

38+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

40− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

40+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

42− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

42+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

44− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

44+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" that"},"finish_reason":null}]}

46− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" given"},"finish_reason":null}]}

46+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

48− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" request"},"finish_reason":null}]}

48+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

50− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

50+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Title"},"finish_reason":null}]}

52− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" The"},"finish_reason":null}]}

52+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" should"},"finish_reason":null}]}

54− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

54+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

56− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

56+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

58− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

58+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" same"},"finish_reason":null}]}

60− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"fetch"},"finish_reason":null}]}

60+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" language"},"finish_reason":null}]}

62− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

62+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

64− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

64+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"English"},"finish_reason":null}]}

66− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" from"},"finish_reason":null}]}

66+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")."},"finish_reason":null}]}

68− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" https"},"finish_reason":null}]}

68+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Not"},"finish_reason":null}]}

70− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"://"},"finish_reason":null}]}

70+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" more"},"finish_reason":null}]}

72− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"example"},"finish_reason":null}]}

72+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" than"},"finish_reason":null}]}

74− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"-files"},"finish_reason":null}]}

74+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

76− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".online"},"finish_reason":null}]}

76+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

78− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"-con"},"finish_reason":null}]}

78+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

80− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"vert"},"finish_reason":null}]}

80+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

82− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".com"},"finish_reason":null}]}

82+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Summ"},"finish_reason":null}]}

84− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"/"},"finish_reason":null}]}

84+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ar"},"finish_reason":null}]}

86− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"website"},"finish_reason":null}]}

86+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ize"},"finish_reason":null}]}

88− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"/html"},"finish_reason":null}]}

88+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

90− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"/example"},"finish_reason":null}]}

90+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

92− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".html"},"finish_reason":null}]}

92+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Check"},"finish_reason":null}]}

94− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

94+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" if"},"finish_reason":null}]}

96− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" tell"},"finish_reason":null}]}

96+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" example"},"finish_reason":null}]}

98− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" me"},"finish_reason":null}]}

98+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".html"},"finish_reason":null}]}

100− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" if"},"finish_reason":null}]}

100+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" contains"},"finish_reason":null}]}

101

102− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" it"},"finish_reason":null}]}

102+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" '"},"finish_reason":null}]}

103

104− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" contains"},"finish_reason":null}]}

104+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"John"},"finish_reason":null}]}

105

106− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

106+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Doe"},"finish_reason":null}]}

107

108− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" word"},"finish_reason":null}]}

108+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"'"},"finish_reason":null}]}

109

110− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" '"},"finish_reason":null}]}

110+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

111

112− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"John"},"finish_reason":null}]}

112+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That"},"finish_reason":null}]}

113

114− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Doe"},"finish_reason":null}]}

114+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" fits"},"finish_reason":null}]}

115

116− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"'"},"finish_reason":null}]}

116+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

117

118− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

118+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Ensure"},"finish_reason":null}]}

119

120− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Title"},"finish_reason":null}]}

120+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" under"},"finish_reason":null}]}

121

122− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" should"},"finish_reason":null}]}

122+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

123

124− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" summarize"},"finish_reason":null}]}

124+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

125

126− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

126+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

127

128− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Title"},"finish_reason":null}]}

128+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

129

130− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

130+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

131

132− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" English"},"finish_reason":null}]}

132+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

133

134− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

134+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

135

136− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" under"},"finish_reason":null}]}

136+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Check"},"finish_reason":null}]}

137

138− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

138+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" if"},"finish_reason":null}]}

139

140− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

140+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" example"},"finish_reason":null}]}

141

142− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

142+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".html"},"finish_reason":null}]}

143

144− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

144+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" contains"},"finish_reason":null}]}

145

146− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Something"},"finish_reason":null}]}

146+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" '"},"finish_reason":null}]}

147

148− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" like"},"finish_reason":null}]}

148+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"John"},"finish_reason":null}]}

149

150− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

150+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Doe"},"finish_reason":null}]}

151

152− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Check"},"finish_reason":null}]}

152+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"'\""},"finish_reason":null}]}

153

154− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" if"},"finish_reason":null}]}

154+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" length"},"finish_reason":null}]}

155

156− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" example"},"finish_reason":null}]}

156+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

157

158− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".html"},"finish_reason":null}]}

158+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Let's"},"finish_reason":null}]}

159

160− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" contains"},"finish_reason":null}]}

160+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

161

162− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" '"},"finish_reason":null}]}

162+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

163

164− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"John"},"finish_reason":null}]}

164+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

165

166− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Doe"},"finish_reason":null}]}

166+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Check"},"finish_reason":null}]}

167

168− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"'"},"finish_reason":null}]}

168+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

169

170− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

170+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"5"},"finish_reason":null}]}

171

172− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

172+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" +"},"finish_reason":null}]}

173

174− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

174+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

175

176− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

176+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

177

178− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

178+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

179

180− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Check"},"finish_reason":null}]}

180+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"6"},"finish_reason":null}]}

181

182− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" if"},"finish_reason":null}]}

182+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

183

184− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" example"},"finish_reason":null}]}

184+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

185

186− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".html"},"finish_reason":null}]}

186+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"if"},"finish_reason":null}]}

187

188− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" contains"},"finish_reason":null}]}

188+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

189

190− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" '"},"finish_reason":null}]}

190+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"2"},"finish_reason":null}]}

191

192− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"John"},"finish_reason":null}]}

192+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

193

194− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Doe"},"finish_reason":null}]}

194+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"8"},"finish_reason":null}]}

195

196− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"'\""},"finish_reason":null}]}

196+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

197

198− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" -\u003e"},"finish_reason":null}]}

198+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

199

200− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Let's"},"finish_reason":null}]}

200+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

201

202− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

202+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"9"},"finish_reason":null}]}

203

204− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

204+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

205

206− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" C"},"finish_reason":null}]}

206+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

207

208− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

208+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"example"},"finish_reason":null}]}

209

210− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

210+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".html"},"finish_reason":null}]}

211

212− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

212+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

213

214− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"h"},"finish_reason":null}]}

214+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"12"},"finish_reason":null}]}

215

216− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"2"},"finish_reason":null}]}

216+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

217

218− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"e"},"finish_reason":null}]}

218+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"21"},"finish_reason":null}]}

219

220− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"3"},"finish_reason":null}]}

220+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

221

222− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"c"},"finish_reason":null}]}

222+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

223

224− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

224+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

225

226− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"k"},"finish_reason":null}]}

226+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"22"},"finish_reason":null}]}

227

228− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"5"},"finish_reason":null}]}

228+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

229

230− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

230+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

231

232− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

232+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"contains"},"finish_reason":null}]}

233

234− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

234+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

235

236− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"6"},"finish_reason":null}]}

236+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"8"},"finish_reason":null}]}

237

238− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"i"},"finish_reason":null}]}

238+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

239

240− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"7"},"finish_reason":null}]}

240+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"30"},"finish_reason":null}]}

241

242− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"f"},"finish_reason":null}]}

242+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

243

244− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"8"},"finish_reason":null}]}

244+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

245

246− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

246+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

247

248− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

248+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"31"},"finish_reason":null}]}

249

250− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

250+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

251

252− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"9"},"finish_reason":null}]}

252+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \"'"},"finish_reason":null}]}

253

254− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"e"},"finish_reason":null}]}

254+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"John"},"finish_reason":null}]}

255

256− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"10"},"finish_reason":null}]}

256+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

257

258− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"x"},"finish_reason":null}]}

258+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"5"},"finish_reason":null}]}

259

260− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"11"},"finish_reason":null}]}

260+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

261

262− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"a"},"finish_reason":null}]}

262+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"36"},"finish_reason":null}]}

263

264− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"12"},"finish_reason":null}]}

264+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

265

266− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"m"},"finish_reason":null}]}

266+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

267

268− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"13"},"finish_reason":null}]}

268+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

269

270− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"p"},"finish_reason":null}]}

270+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

271

272− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"14"},"finish_reason":null}]}

272+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

273

274− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"l"},"finish_reason":null}]}

274+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"37"},"finish_reason":null}]}

275

276− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"15"},"finish_reason":null}]}

276+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

277

278− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"e"},"finish_reason":null}]}

278+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

279

280− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"16"},"finish_reason":null}]}

280+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Doe"},"finish_reason":null}]}

281

282− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

282+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"'\""},"finish_reason":null}]}

283

284− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"17"},"finish_reason":null}]}

284+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

285

286− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"h"},"finish_reason":null}]}

286+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

287

288− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"18"},"finish_reason":null}]}

288+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"41"},"finish_reason":null}]}

289

290− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"t"},"finish_reason":null}]}

290+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

291

292− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"19"},"finish_reason":null}]}

292+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Total"},"finish_reason":null}]}

293

294− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"m"},"finish_reason":null}]}

294+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

295

296− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"20"},"finish_reason":null}]}

296+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"41"},"finish_reason":null}]}

297

298− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"l"},"finish_reason":null}]}

298+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

299

300− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"21"},"finish_reason":null}]}

300+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

301

302− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

302+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Good"},"finish_reason":null}]}

303

304− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

304+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".\n\n"},"finish_reason":null}]}

305

306− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

306+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Return"},"finish_reason":null}]}

307

308− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"22"},"finish_reason":null}]}

308+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" just"},"finish_reason":null}]}

309

310− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"c"},"finish_reason":null}]}

310+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" that"},"finish_reason":null}]}

311

312− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"23"},"finish_reason":null}]}

312+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

313

314− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"o"},"finish_reason":null}]}

314+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

315

316− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"24"},"finish_reason":null}]}

316+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Check"},"finish_reason":null}]}

317

318− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"n"},"finish_reason":null}]}

318+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" if"},"finish_reason":null}]}

319

320− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"25"},"finish_reason":null}]}

320+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" example"},"finish_reason":null}]}

321

322− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"t"},"finish_reason":null}]}

322+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":".html"},"finish_reason":null}]}

323

324− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"26"},"finish_reason":null}]}

324+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" contains"},"finish_reason":null}]}

325

326− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"a"},"finish_reason":null}]}

326+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" John"},"finish_reason":null}]}

327

328− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"27"},"finish_reason":null}]}

328+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" Doe"},"finish_reason":null}]}

329

330− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"i"},"finish_reason":null}]}

330+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

331

332− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"28"},"finish_reason":null}]}

333−

334− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"n"},"finish_reason":null}]}

335−

336− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"29"},"finish_reason":null}]}

337−

338− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"s"},"finish_reason":null}]}

339−

340− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"30"},"finish_reason":null}]}

341−

342− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

343−

344− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

345−

346− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

347−

348− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"31"},"finish_reason":null}]}

349−

350− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"'"},"finish_reason":null}]}

351−

352− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"32"},"finish_reason":null}]}

353−

354− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"J"},"finish_reason":null}]}

355−

356− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"33"},"finish_reason":null}]}

357−

358− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"o"},"finish_reason":null}]}

359−

360− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"34"},"finish_reason":null}]}

361−

362− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"h"},"finish_reason":null}]}

363−

364− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"35"},"finish_reason":null}]}

365−

366− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"n"},"finish_reason":null}]}

367−

368− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"36"},"finish_reason":null}]}

369−

370− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

371−

372− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

373−

374− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

375−

376− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"37"},"finish_reason":null}]}

377−

378− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"D"},"finish_reason":null}]}

379−

380− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"38"},"finish_reason":null}]}

381−

382− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"o"},"finish_reason":null}]}

383−

384− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"39"},"finish_reason":null}]}

385−

386− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"e"},"finish_reason":null}]}

387−

388− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"40"},"finish_reason":null}]}

389−

390− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"'"},"finish_reason":null}]}

391−

392− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" -\u003e"},"finish_reason":null}]}

393−

394− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

395−

396− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"40"},"finish_reason":null}]}

397−

398− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

399−

400− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

401−

402− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Good"},"finish_reason":null}]}

403−

404− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".\n\n"},"finish_reason":null}]}

405−

406− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Return"},"finish_reason":null}]}

407−

408− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" that"},"finish_reason":null}]}

409−

410− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

411−

412− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" only"},"finish_reason":null}]}

413−

414− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

415−

416− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Check"},"finish_reason":null}]}

417−

418− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" if"},"finish_reason":null}]}

419−

420− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" example"},"finish_reason":null}]}

421−

422− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":".html"},"finish_reason":null}]}

423−

424− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" contains"},"finish_reason":null}]}

425−

426− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" '"},"finish_reason":null}]}

427−

428− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"John"},"finish_reason":null}]}

429−

430− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" Doe"},"finish_reason":null}]}

431−

432− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"'"},"finish_reason":null}]}

433−

434− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

435−

436− data: {"id":"chatcmpl-b9f0857fe47425c2e33e4532","object":"chat.completion.chunk","created":1779218880,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":165,"completion_tokens":212,"total_tokens":441,"prompt_tokens_details":{"cached_tokens":64},"completion_tokens_details":{"reasoning_tokens":193},"cost":{"usd":0.00017850000000000003,"hypercredits":0.0035700000000000003},"remaining":{"hypercredits":0}}}

332+ data: {"id":"chatcmpl-3b28ebbfe079f5813ad4baba","object":"chat.completion.chunk","created":1780069685,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":232,"completion_tokens":160,"total_tokens":392,"completion_tokens_details":{"reasoning_tokens":144},"cost":{"usd":0.0002272,"hypercredits":0.004543999999999999},"remaining":{"hypercredits":0}}}

437

333

438 data: [DONE]

334 data: [DONE]

439

335

@@ -442,15 +338,15 @@ interactions:

442 - text/event-stream

338 - text/event-stream

443 status: 200 OK

339 status: 200 OK

444 code: 200

340 code: 200

445− duration: 505.595667ms

341+ duration: 564.254458ms

446- id: 1

342- id: 1

447 request:

343 request:

448 proto: HTTP/1.1

344 proto: HTTP/1.1

449 proto_major: 1

345 proto_major: 1

450 proto_minor: 1

346 proto_minor: 1

451− content_length: 34047

347+ content_length: 34028

452 host: ""

348 host: ""

453− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/fetch_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"fetch the content from https://example-files.online-convert.com/website/html/example.html and tell me if it contains the word 'John Doe'\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

349+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/fetch_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"fetch the content from https://example-files.online-convert.com/website/html/example.html and tell me if it contains the word 'John Doe'\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

454 headers:

350 headers:

455 Accept:

351 Accept:

456 - application/json

352 - application/json

@@ -467,61 +363,93 @@ interactions:

467 content_length: -1

363 content_length: -1

468 uncompressed: true

364 uncompressed: true

469 body: |+

365 body: |+

470− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

366+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

367+

368+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

369+

370+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

371+

372+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" wants"},"finish_reason":null}]}

471

373

472− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

374+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" me"},"finish_reason":null}]}

473

375

474− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

376+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

475

377

476− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" wants me to"},"finish_reason":null}]}

378+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" fetch"},"finish_reason":null}]}

477

379

478− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" fetch content"},"finish_reason":null}]}

380+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

479

381

480− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" from a"},"finish_reason":null}]}

382+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" from"},"finish_reason":null}]}

481

383

482− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" URL and check"},"finish_reason":null}]}

384+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

483

385

484− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" if it contains"},"finish_reason":null}]}

386+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" URL"},"finish_reason":null}]}

485

387

486− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" the word"},"finish_reason":null}]}

388+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

487

389

488− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" '"},"finish_reason":null}]}

390+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" check"},"finish_reason":null}]}

489

391

490− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"John Doe"},"finish_reason":null}]}

392+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" if"},"finish_reason":null}]}

491

393

492− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"'."},"finish_reason":null}]}

394+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" it"},"finish_reason":null}]}

493

395

494− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_30823fe8613e4fd099a524eb","type":"function","index":0,"function":{"name":"fetch","arguments":""}}]},"finish_reason":null}]}

396+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" contains"},"finish_reason":null}]}

495

397

496− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

398+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

497

399

498− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

400+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" word"},"finish_reason":null}]}

499

401

500− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"format\": "}}]},"finish_reason":null}]}

402+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" '"},"finish_reason":null}]}

501

403

502− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"text\", "}}]},"finish_reason":null}]}

404+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"John"},"finish_reason":null}]}

503

405

504− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"url\": "}}]},"finish_reason":null}]}

406+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" Doe"},"finish_reason":null}]}

505

407

506− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"https://example"}}]},"finish_reason":null}]}

408+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"'."},"finish_reason":null}]}

507

409

508− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-files"}}]},"finish_reason":null}]}

410+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_2TxDxIRLeDX97ZGITAVav57g","type":"function","index":0,"function":{"name":"fetch","arguments":""}}]},"finish_reason":null}]}

509

411

510− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".online"}}]},"finish_reason":null}]}

412+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

511

413

512− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-convert.com"}}]},"finish_reason":null}]}

414+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

513

415

514− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/website"}}]},"finish_reason":null}]}

416+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"format\": "}}]},"finish_reason":null}]}

515

417

516− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/html/example"}}]},"finish_reason":null}]}

418+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"text"}}]},"finish_reason":null}]}

517

419

518− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".html\""}}]},"finish_reason":null}]}

420+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\", \"url\": "}}]},"finish_reason":null}]}

519

421

520− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"}"}}]},"finish_reason":null}]}

422+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"https"}}]},"finish_reason":null}]}

521

423

522− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

424+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"://"}}]},"finish_reason":null}]}

523

425

426+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"example"}}]},"finish_reason":null}]}

427+

428+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-files"}}]},"finish_reason":null}]}

429+

430+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".online"}}]},"finish_reason":null}]}

431+

432+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-"}}]},"finish_reason":null}]}

433+

434+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"convert"}}]},"finish_reason":null}]}

435+

436+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".com"}}]},"finish_reason":null}]}

437+

438+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/"}}]},"finish_reason":null}]}

439+

440+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"website"}}]},"finish_reason":null}]}

441+

442+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/html"}}]},"finish_reason":null}]}

443+

444+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/example"}}]},"finish_reason":null}]}

445+

446+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".html"}}]},"finish_reason":null}]}

447+

448+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"}"}}]},"finish_reason":null}]}

449+

450+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

451+

524− data: {"id":"chatcmpl-8284161da418267710430452","object":"chat.completion.chunk","created":1779218882,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":730,"completion_tokens":50,"total_tokens":7884,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":21},"cost":{"usd":0.0014074000000000003,"hypercredits":0.028148000000000003},"remaining":{"hypercredits":0}}}

452+ data: {"id":"chatcmpl-571ae68110255a47c63f8340","object":"chat.completion.chunk","created":1780069685,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":725,"completion_tokens":50,"total_tokens":7879,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":23},"cost":{"usd":0.00125635,"hypercredits":0.025127},"remaining":{"hypercredits":0}}}

525

453

526 data: [DONE]

454 data: [DONE]

527

455

@@ -530,7 +458,7 @@ interactions:

530 - text/event-stream

458 - text/event-stream

531 status: 200 OK

459 status: 200 OK

532 code: 200

460 code: 200

533− duration: 2.702715125s

461+ duration: 550.064ms

534- id: 2

462- id: 2

535 request:

463 request:

536 proto: HTTP/1.1

464 proto: HTTP/1.1

@@ -601,15 +529,15 @@ interactions:

601 - text/html; charset=UTF-8

529 - text/html; charset=UTF-8

602 status: 200 OK

530 status: 200 OK

603 code: 200

531 code: 200

604− duration: 156.879917ms

532+ duration: 146.217209ms

605- id: 3

533- id: 3

606 request:

534 request:

607 proto: HTTP/1.1

535 proto: HTTP/1.1

608 proto_major: 1

536 proto_major: 1

609 proto_minor: 1

537 proto_minor: 1

610− content_length: 34579

538+ content_length: 34560

611 host: ""

539 host: ""

612− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/fetch_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"fetch the content from https://example-files.online-convert.com/website/html/example.html and tell me if it contains the word 'John Doe'\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_30823fe8613e4fd099a524eb\",\"function\":{\"arguments\":\"{\\\"format\\\": \\\"text\\\", \\\"url\\\": \\\"https://example-files.online-convert.com/website/html/example.html\\\"}\",\"name\":\"fetch\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"The user wants me to fetch content from a URL and check if it contains the word 'John Doe'.\"},{\"content\":\"HTML test file Purpose: Provide example of this file type Document file type: HTML Version: 1.0 Remark:\",\"tool_call_id\":\"call_30823fe8613e4fd099a524eb\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

540+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/fetch_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"fetch the content from https://example-files.online-convert.com/website/html/example.html and tell me if it contains the word 'John Doe'\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_2TxDxIRLeDX97ZGITAVav57g\",\"function\":{\"arguments\":\"{\\\"format\\\": \\\"text\\\", \\\"url\\\": \\\"https://example-files.online-convert.com/website/html/example.html\\\"}\",\"name\":\"fetch\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"The user wants me to fetch content from a URL and check if it contains the word 'John Doe'.\"},{\"content\":\"HTML test file Purpose: Provide example of this file type Document file type: HTML Version: 1.0 Remark:\",\"tool_call_id\":\"call_2TxDxIRLeDX97ZGITAVav57g\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

613 headers:

541 headers:

614 Accept:

542 Accept:

615 - application/json

543 - application/json

@@ -626,29 +554,39 @@ interactions:

626 content_length: -1

554 content_length: -1

627 uncompressed: true

555 uncompressed: true

628 body: |+

556 body: |+

629− data: {"id":"chatcmpl-97436907762babde085a7316","object":"chat.completion.chunk","created":1779218885,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

557+ data: {"id":"chatcmpl-6818662871afdceed8cba80a","object":"chat.completion.chunk","created":1780069689,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

558+

559+ data: {"id":"chatcmpl-6818662871afdceed8cba80a","object":"chat.completion.chunk","created":1780069689,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"No"},"finish_reason":null}]}

560+

561+ data: {"id":"chatcmpl-6818662871afdceed8cba80a","object":"chat.completion.chunk","created":1780069689,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":","},"finish_reason":null}]}

562+

563+ data: {"id":"chatcmpl-6818662871afdceed8cba80a","object":"chat.completion.chunk","created":1780069689,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" the"},"finish_reason":null}]}

564+

565+ data: {"id":"chatcmpl-6818662871afdceed8cba80a","object":"chat.completion.chunk","created":1780069689,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" fetched"},"finish_reason":null}]}

566+

567+ data: {"id":"chatcmpl-6818662871afdceed8cba80a","object":"chat.completion.chunk","created":1780069689,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" content"},"finish_reason":null}]}

630

568

631− data: {"id":"chatcmpl-97436907762babde085a7316","object":"chat.completion.chunk","created":1779218885,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"The"},"finish_reason":null}]}

569+ data: {"id":"chatcmpl-6818662871afdceed8cba80a","object":"chat.completion.chunk","created":1780069689,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" does"},"finish_reason":null}]}

632

570

633− data: {"id":"chatcmpl-97436907762babde085a7316","object":"chat.completion.chunk","created":1779218885,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" fetched"},"finish_reason":null}]}

571+ data: {"id":"chatcmpl-6818662871afdceed8cba80a","object":"chat.completion.chunk","created":1780069689,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" not"},"finish_reason":null}]}

634

572

635− data: {"id":"chatcmpl-97436907762babde085a7316","object":"chat.completion.chunk","created":1779218885,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" content does **"},"finish_reason":null}]}

573+ data: {"id":"chatcmpl-6818662871afdceed8cba80a","object":"chat.completion.chunk","created":1780069689,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" contain"},"finish_reason":null}]}

636

574

637− data: {"id":"chatcmpl-97436907762babde085a7316","object":"chat.completion.chunk","created":1779218885,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"not"},"finish_reason":null}]}

575+ data: {"id":"chatcmpl-6818662871afdceed8cba80a","object":"chat.completion.chunk","created":1780069689,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" the"},"finish_reason":null}]}

638

576

639− data: {"id":"chatcmpl-97436907762babde085a7316","object":"chat.completion.chunk","created":1779218885,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"** contain"},"finish_reason":null}]}

577+ data: {"id":"chatcmpl-6818662871afdceed8cba80a","object":"chat.completion.chunk","created":1780069689,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" word"},"finish_reason":null}]}

640

578

641− data: {"id":"chatcmpl-97436907762babde085a7316","object":"chat.completion.chunk","created":1779218885,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" the word"},"finish_reason":null}]}

579+ data: {"id":"chatcmpl-6818662871afdceed8cba80a","object":"chat.completion.chunk","created":1780069689,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" \""},"finish_reason":null}]}

642

580

643− data: {"id":"chatcmpl-97436907762babde085a7316","object":"chat.completion.chunk","created":1779218885,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" '"},"finish_reason":null}]}

581+ data: {"id":"chatcmpl-6818662871afdceed8cba80a","object":"chat.completion.chunk","created":1780069689,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"John"},"finish_reason":null}]}

644

582

645− data: {"id":"chatcmpl-97436907762babde085a7316","object":"chat.completion.chunk","created":1779218885,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"John Doe"},"finish_reason":null}]}

583+ data: {"id":"chatcmpl-6818662871afdceed8cba80a","object":"chat.completion.chunk","created":1780069689,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" Doe"},"finish_reason":null}]}

646

584

647− data: {"id":"chatcmpl-97436907762babde085a7316","object":"chat.completion.chunk","created":1779218885,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"'."},"finish_reason":null}]}

585+ data: {"id":"chatcmpl-6818662871afdceed8cba80a","object":"chat.completion.chunk","created":1780069689,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"\"."},"finish_reason":null}]}

648

586

649− data: {"id":"chatcmpl-97436907762babde085a7316","object":"chat.completion.chunk","created":1779218885,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

587+ data: {"id":"chatcmpl-6818662871afdceed8cba80a","object":"chat.completion.chunk","created":1780069689,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

650

588

651− data: {"id":"chatcmpl-97436907762babde085a7316","object":"chat.completion.chunk","created":1779218885,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":83,"completion_tokens":16,"total_tokens":7907,"prompt_tokens_details":{"cached_tokens":7808},"completion_tokens_details":{"reasoning_tokens":15},"cost":{"usd":0.0002609,"hypercredits":0.0052179999999999995},"remaining":{"hypercredits":0}}}

589+ data: {"id":"chatcmpl-6818662871afdceed8cba80a","object":"chat.completion.chunk","created":1780069689,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":78,"completion_tokens":16,"total_tokens":7902,"prompt_tokens_details":{"cached_tokens":7808},"cost":{"usd":0.00016938,"hypercredits":0.0033876},"remaining":{"hypercredits":0}}}

652

590

653 data: [DONE]

591 data: [DONE]

654

592

@@ -657,4 +595,4 @@ interactions:

657 - text/event-stream

595 - text/event-stream

658 status: 200 OK

596 status: 200 OK

659 code: 200

597 code: 200

660− duration: 1.924023708s

598+ duration: 2.6751105s

internal/agent/testdata/TestCoderAgent/glm-5.1/glob_tool.yaml+487−289

@@ -2,6 +2,98 @@

2version: 2

3interactions:

4- id: 0

5+ request:

6+ proto: HTTP/1.1

7+ proto_major: 1

8+ proto_minor: 1

9+ content_length: 33946

10+ host: ""

11+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/glob_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use glob to find all .go files in the current directory\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

12+ headers:

13+ Accept:

14+ - application/json

15+ Content-Type:

16+ - application/json

17+ User-Agent:

18+ - Charm-Crush/devel (https://charm.land/crush)

19+ url: https://hyper.charm.land/v1/chat/completions

20+ method: POST

21+ response:

22+ proto: HTTP/2.0

23+ proto_major: 2

24+ proto_minor: 0

25+ content_length: -1

26+ uncompressed: true

27+ body: |+

28+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

29+

30+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

31+

32+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

33+

34+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" wants"},"finish_reason":null}]}

35+

36+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" me"},"finish_reason":null}]}

37+

38+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

39+

40+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" find"},"finish_reason":null}]}

41+

42+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" all"},"finish_reason":null}]}

43+

44+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

45+

46+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"go"},"finish_reason":null}]}

47+

48+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

49+

50+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

51+

52+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

53+

54+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" current"},"finish_reason":null}]}

55+

56+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" directory"},"finish_reason":null}]}

57+

58+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" using"},"finish_reason":null}]}

59+

60+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

61+

62+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" glob"},"finish_reason":null}]}

63+

64+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" tool"},"finish_reason":null}]}

65+

66+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

67+

68+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_fl8kSELeCV44nzdl56WTZKve","type":"function","index":0,"function":{"name":"glob","arguments":""}}]},"finish_reason":null}]}

69+

70+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

71+

72+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

73+

74+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"pattern\": "}}]},"finish_reason":null}]}

75+

76+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"**"}}]},"finish_reason":null}]}

77+

78+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/*."}}]},"finish_reason":null}]}

79+

80+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"go"}}]},"finish_reason":null}]}

81+

82+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"}"}}]},"finish_reason":null}]}

83+

84+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

85+

86+ data: {"id":"chatcmpl-a80e274f5ef31a4f7f4a0757","object":"chat.completion.chunk","created":1780069690,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":708,"completion_tokens":32,"total_tokens":7844,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":22},"cost":{"usd":0.0011542800000000001,"hypercredits":0.0230856},"remaining":{"hypercredits":0}}}

87+

88+ data: [DONE]

89+

90+ headers:

91+ Content-Type:

92+ - text/event-stream

93+ status: 200 OK

94+ code: 200

95+ duration: 625.636708ms

96+- id: 1

5 request:

97 request:

6 proto: HTTP/1.1

98 proto: HTTP/1.1

7 proto_major: 1

99 proto_major: 1

@@ -25,525 +117,621 @@ interactions:

25 content_length: -1

117 content_length: -1

26 uncompressed: true

118 uncompressed: true

27 body: |+

119 body: |+

28− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

120+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

121

30− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"We"},"finish_reason":null}]}

122+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

123

32− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" need"},"finish_reason":null}]}

124+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

125

34− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

126+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" says"},"finish_reason":null}]}

127

36− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" produce"},"finish_reason":null}]}

128+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

129

38− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

130+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

131

40− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" short"},"finish_reason":null}]}

132+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Generate"},"finish_reason":null}]}

133

42− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

134+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

135

44− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

136+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

137

46− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" max"},"finish_reason":null}]}

138+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

139

48− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

140+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

141

50− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

142+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

143

52− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

144+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" following"},"finish_reason":null}]}

145

54− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

146+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

147

56− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

148+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":\n\n"},"finish_reason":null}]}

149

58− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" same"},"finish_reason":null}]}

150+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"use"},"finish_reason":null}]}

151

60− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" language"},"finish_reason":null}]}

152+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" glob"},"finish_reason":null}]}

153

62− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

154+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

155

64− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"English"},"finish_reason":null}]}

156+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" find"},"finish_reason":null}]}

157

66− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")."},"finish_reason":null}]}

158+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" all"},"finish_reason":null}]}

159

68− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Summ"},"finish_reason":null}]}

160+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

161

70− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ar"},"finish_reason":null}]}

162+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"go"},"finish_reason":null}]}

163

72− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ize"},"finish_reason":null}]}

164+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

165

74− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

166+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

167

76− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

168+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

169

78− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"use"},"finish_reason":null}]}

170+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" current"},"finish_reason":null}]}

171

80− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" glob"},"finish_reason":null}]}

172+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" directory"},"finish_reason":null}]}

173

82− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

174+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

175

84− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" find"},"finish_reason":null}]}

176+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" So"},"finish_reason":null}]}

177

86− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" all"},"finish_reason":null}]}

178+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" need"},"finish_reason":null}]}

179

88− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

180+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

181

90− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"go"},"finish_reason":null}]}

182+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

183

92− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

184+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" same"},"finish_reason":null}]}

185

94− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

186+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" language"},"finish_reason":null}]}

187

96− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

188+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

189

98− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" current"},"finish_reason":null}]}

190+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"English"},"finish_reason":null}]}

191

100− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" directory"},"finish_reason":null}]}

192+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")."},"finish_reason":null}]}

101

193

102− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

194+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Under"},"finish_reason":null}]}

103

195

104− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Title"},"finish_reason":null}]}

196+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

105

197

106− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" could"},"finish_reason":null}]}

198+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

107

199

108− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

200+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

109

201

110− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

202+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

111

203

112− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Find"},"finish_reason":null}]}

204+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Summ"},"finish_reason":null}]}

113

205

114− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" all"},"finish_reason":null}]}

206+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ar"},"finish_reason":null}]}

115

207

116− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

208+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ize"},"finish_reason":null}]}

117

209

118− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"go"},"finish_reason":null}]}

210+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

119

211

120− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

212+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Something"},"finish_reason":null}]}

121

213

122− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

214+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" like"},"finish_reason":null}]}

123

215

124− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" glob"},"finish_reason":null}]}

216+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

125

217

126− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

218+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Using"},"finish_reason":null}]}

127

219

128− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" current"},"finish_reason":null}]}

220+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" glob"},"finish_reason":null}]}

129

221

130− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" directory"},"finish_reason":null}]}

222+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

131

223

132− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

224+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" locate"},"finish_reason":null}]}

133

225

134− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

226+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

135

227

136− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" under"},"finish_reason":null}]}

228+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"go"},"finish_reason":null}]}

137

229

138− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

230+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

139

231

140− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

232+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

141

233

142− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

234+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" current"},"finish_reason":null}]}

143

235

144− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

236+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" directory"},"finish_reason":null}]}

145

237

146− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Let's"},"finish_reason":null}]}

238+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

147

239

148− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

240+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

149

241

150− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

242+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" under"},"finish_reason":null}]}

151

243

152− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

244+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

153

245

154− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Find"},"finish_reason":null}]}

246+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

155

247

156− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" all"},"finish_reason":null}]}

248+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

157

249

158− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

250+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"?"},"finish_reason":null}]}

159

251

160− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"go"},"finish_reason":null}]}

252+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

161

253

162− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

254+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

163

255

164− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

256+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

165

257

166− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" glob"},"finish_reason":null}]}

258+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Using"},"finish_reason":null}]}

167

259

168− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

260+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" glob"},"finish_reason":null}]}

169

261

170− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" current"},"finish_reason":null}]}

262+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

171

263

172− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" directory"},"finish_reason":null}]}

264+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" locate"},"finish_reason":null}]}

173

265

174− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

266+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

175

267

176− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" length"},"finish_reason":null}]}

268+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"go"},"finish_reason":null}]}

177

269

178− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

270+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

179

271

180− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

272+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

181

273

182− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

274+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" current"},"finish_reason":null}]}

183

275

184− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"?"},"finish_reason":null}]}

276+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" directory"},"finish_reason":null}]}

185

277

186− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Let's"},"finish_reason":null}]}

278+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

187

279

188− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

280+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" length"},"finish_reason":null}]}

189

281

190− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

282+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

191

283

192− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" F"},"finish_reason":null}]}

284+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" let's"},"finish_reason":null}]}

193

285

194− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

286+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

195

287

196− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

288+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

197

289

198− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")i"},"finish_reason":null}]}

290+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

199

291

200− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"2"},"finish_reason":null}]}

292+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Using"},"finish_reason":null}]}

201

293

202− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"n"},"finish_reason":null}]}

294+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

203

295

204− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"3"},"finish_reason":null}]}

296+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"5"},"finish_reason":null}]}

205

297

206− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"d"},"finish_reason":null}]}

298+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

207

299

208− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

300+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" +"},"finish_reason":null}]}

209

301

210− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

302+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

211

303

212− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

304+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

213

305

214− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

306+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ="},"finish_reason":null}]}

215

307

216− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"5"},"finish_reason":null}]}

308+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"6"},"finish_reason":null}]}

217

309

218− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"a"},"finish_reason":null}]}

310+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

219

311

220− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"6"},"finish_reason":null}]}

312+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" glob"},"finish_reason":null}]}

221

313

222− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"l"},"finish_reason":null}]}

314+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

223

315

224− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"7"},"finish_reason":null}]}

316+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

225

317

226− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"l"},"finish_reason":null}]}

318+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

227

319

228− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"8"},"finish_reason":null}]}

320+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" =\u003e"},"finish_reason":null}]}

229

321

230− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

322+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"10"},"finish_reason":null}]}

231

323

232− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

324+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

233

325

234− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

326+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

235

327

236− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"9"},"finish_reason":null}]}

328+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

237

329

238− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

330+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"11"},"finish_reason":null}]}

239

331

240− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"10"},"finish_reason":null}]}

332+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

241

333

242− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"g"},"finish_reason":null}]}

334+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

243

335

244− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"11"},"finish_reason":null}]}

336+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

245

337

246− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"o"},"finish_reason":null}]}

338+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"2"},"finish_reason":null}]}

247

339

248− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"12"},"finish_reason":null}]}

340+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")=\u003e"},"finish_reason":null}]}

249

341

250− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

342+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"13"},"finish_reason":null}]}

251

343

252− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

344+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

253

345

254− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

346+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

255

347

256− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"13"},"finish_reason":null}]}

348+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

257

349

258− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"f"},"finish_reason":null}]}

350+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"14"},"finish_reason":null}]}

259

351

260− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"14"},"finish_reason":null}]}

352+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

261

353

262− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"i"},"finish_reason":null}]}

354+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" locate"},"finish_reason":null}]}

263

355

264− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"15"},"finish_reason":null}]}

356+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

265

357

266− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"l"},"finish_reason":null}]}

358+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"6"},"finish_reason":null}]}

267

359

268− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"16"},"finish_reason":null}]}

360+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")=\u003e"},"finish_reason":null}]}

269

361

270− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"e"},"finish_reason":null}]}

362+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"20"},"finish_reason":null}]}

271

363

272− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"17"},"finish_reason":null}]}

364+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

273

365

274− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"s"},"finish_reason":null}]}

366+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

275

367

276− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"18"},"finish_reason":null}]}

368+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

277

369

278− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

370+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"21"},"finish_reason":null}]}

279

371

280− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

372+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

281

373

282− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

374+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

283

375

284− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"19"},"finish_reason":null}]}

376+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"go"},"finish_reason":null}]}

285

377

286− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"w"},"finish_reason":null}]}

378+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

287

379

288− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"20"},"finish_reason":null}]}

380+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"3"},"finish_reason":null}]}

289

381

290− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"i"},"finish_reason":null}]}

382+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")=\u003e"},"finish_reason":null}]}

291

383

292− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"21"},"finish_reason":null}]}

384+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"24"},"finish_reason":null}]}

293

385

294− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"t"},"finish_reason":null}]}

386+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

295

387

296− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"22"},"finish_reason":null}]}

388+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

297

389

298− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"h"},"finish_reason":null}]}

390+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

299

391

300− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"23"},"finish_reason":null}]}

392+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"25"},"finish_reason":null}]}

301

393

302− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

394+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

303

395

304− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

396+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

305

397

306− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

398+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

307

399

308− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"24"},"finish_reason":null}]}

400+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"5"},"finish_reason":null}]}

309

401

310− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"g"},"finish_reason":null}]}

402+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")=\u003e"},"finish_reason":null}]}

311

403

312− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"25"},"finish_reason":null}]}

404+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"30"},"finish_reason":null}]}

313

405

314− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"l"},"finish_reason":null}]}

406+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

315

407

316− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"26"},"finish_reason":null}]}

408+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

317

409

318− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"o"},"finish_reason":null}]}

410+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

319

411

320− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"27"},"finish_reason":null}]}

412+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"31"},"finish_reason":null}]}

321

413

322− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"b"},"finish_reason":null}]}

414+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

323

415

324− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"28"},"finish_reason":null}]}

416+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

325

417

326− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

418+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

327

419

328− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

420+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"2"},"finish_reason":null}]}

329

421

330− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

422+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")=\u003e"},"finish_reason":null}]}

331

423

332− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"29"},"finish_reason":null}]}

424+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"33"},"finish_reason":null}]}

333

425

334− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"i"},"finish_reason":null}]}

426+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

335

427

336− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"30"},"finish_reason":null}]}

428+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

337

429

338− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"n"},"finish_reason":null}]}

430+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

339

431

340− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"31"},"finish_reason":null}]}

432+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"34"},"finish_reason":null}]}

341

433

342− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

434+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

343

435

344− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

436+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" current"},"finish_reason":null}]}

345

437

346− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

438+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

347

439

348− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"32"},"finish_reason":null}]}

440+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"7"},"finish_reason":null}]}

349

441

350− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"c"},"finish_reason":null}]}

442+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")=\u003e"},"finish_reason":null}]}

351

443

352− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"33"},"finish_reason":null}]}

444+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"41"},"finish_reason":null}]}

353

445

354− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"u"},"finish_reason":null}]}

446+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

355

447

356− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"34"},"finish_reason":null}]}

448+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

357

449

358− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"r"},"finish_reason":null}]}

450+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

359

451

360− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"35"},"finish_reason":null}]}

452+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"42"},"finish_reason":null}]}

361

453

362− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"r"},"finish_reason":null}]}

454+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

363

455

364− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"36"},"finish_reason":null}]}

456+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" directory"},"finish_reason":null}]}

365

457

366− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"e"},"finish_reason":null}]}

458+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

367

459

368− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"37"},"finish_reason":null}]}

460+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"9"},"finish_reason":null}]}

369

461

370− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"n"},"finish_reason":null}]}

462+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")=\u003e"},"finish_reason":null}]}

371

463

372− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"38"},"finish_reason":null}]}

464+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"51"},"finish_reason":null}]}

373

465

374− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"t"},"finish_reason":null}]}

466+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

375

467

376− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"39"},"finish_reason":null}]}

468+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

377

469

378− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

470+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

379

471

380− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

472+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"51"},"finish_reason":null}]}

381

473

382− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

474+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

383

475

384− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"40"},"finish_reason":null}]}

476+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

385

477

386− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"d"},"finish_reason":null}]}

478+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" too"},"finish_reason":null}]}

387

479

388− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"41"},"finish_reason":null}]}

480+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" many"},"finish_reason":null}]}

389

481

390− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"i"},"finish_reason":null}]}

482+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

391

483

392− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"42"},"finish_reason":null}]}

484+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Need"},"finish_reason":null}]}

393

485

394− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"r"},"finish_reason":null}]}

486+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \u003c="},"finish_reason":null}]}

395

487

396− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"43"},"finish_reason":null}]}

488+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

397

489

398− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"e"},"finish_reason":null}]}

490+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

399

491

400− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"44"},"finish_reason":null}]}

492+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Could"},"finish_reason":null}]}

401

493

402− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"c"},"finish_reason":null}]}

494+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" shorten"},"finish_reason":null}]}

403

495

404− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"45"},"finish_reason":null}]}

496+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

405

497

406− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"t"},"finish_reason":null}]}

498+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

407

499

408− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"46"},"finish_reason":null}]}

500+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Glob"},"finish_reason":null}]}

409

501

410− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"o"},"finish_reason":null}]}

502+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

411

503

412− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"47"},"finish_reason":null}]}

504+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" find"},"finish_reason":null}]}

413

505

414− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"r"},"finish_reason":null}]}

506+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

415

507

416− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"48"},"finish_reason":null}]}

508+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"go"},"finish_reason":null}]}

417

509

418− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"y"},"finish_reason":null}]}

510+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

419

511

420− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"49"},"finish_reason":null}]}

512+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

421

513

422− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

514+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" current"},"finish_reason":null}]}

423

515

424− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

516+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" directory"},"finish_reason":null}]}

425

517

426− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

518+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

427

519

428− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"49"},"finish_reason":null}]}

520+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

429

521

430− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

522+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

431

523

432− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

524+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

433

525

434− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" okay"},"finish_reason":null}]}

526+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Glob"},"finish_reason":null}]}

435

527

436− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

528+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"("},"finish_reason":null}]}

437

529

438− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" No"},"finish_reason":null}]}

530+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

439

531

440− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" punctuation"},"finish_reason":null}]}

532+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

441

533

442− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" colon"},"finish_reason":null}]}

534+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" +"},"finish_reason":null}]}

443

535

444− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

536+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

445

537

446− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

538+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

447

539

448− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" fine"},"finish_reason":null}]}

540+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"5"},"finish_reason":null}]}

449

541

450− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

542+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

451

543

452− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Find"},"finish_reason":null}]}

544+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

453

545

454− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" all"},"finish_reason":null}]}

546+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"to"},"finish_reason":null}]}

455

547

456− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" ."},"finish_reason":null}]}

548+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"("},"finish_reason":null}]}

457

549

458− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"go"},"finish_reason":null}]}

550+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"2"},"finish_reason":null}]}

459

551

460− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" files"},"finish_reason":null}]}

552+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

461

553

462− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" with"},"finish_reason":null}]}

554+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"7"},"finish_reason":null}]}

463

555

464− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" glob"},"finish_reason":null}]}

556+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

465

557

466− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" in"},"finish_reason":null}]}

558+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

467

559

468− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" current"},"finish_reason":null}]}

560+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

469

561

470− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" directory"},"finish_reason":null}]}

562+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"8"},"finish_reason":null}]}

471

563

472− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

564+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

473

565

474− data: {"id":"chatcmpl-2cbde74a99dc227b00c2ab64","object":"chat.completion.chunk","created":1779218887,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":148,"completion_tokens":231,"total_tokens":448,"prompt_tokens_details":{"cached_tokens":69},"completion_tokens_details":{"reasoning_tokens":150},"cost":{"usd":0.0001672,"hypercredits":0.0033439999999999998},"remaining":{"hypercredits":0}}}

566+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

475

567

476− data: [DONE]

568+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"find"},"finish_reason":null}]}

477

569

478− headers:

570+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"("},"finish_reason":null}]}

479− Content-Type:

571+

480− - text/event-stream

572+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

481− status: 200 OK

573+

482− code: 200

574+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

483− duration: 1.449573459s

575+

484−- id: 1

576+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"12"},"finish_reason":null}]}

485− request:

577+

486− proto: HTTP/1.1

578+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

487− proto_major: 1

579+

488− proto_minor: 1

580+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

489− content_length: 33965

581+

490− host: ""

582+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

491− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/glob_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use glob to find all .go files in the current directory\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

583+

492− headers:

584+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"13"},"finish_reason":null}]}

493− Accept:

585+

494− - application/json

586+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

495− Content-Type:

587+

496− - application/json

588+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \"."},"finish_reason":null}]}

497− User-Agent:

589+

498− - Charm-Crush/devel (https://charm.land/crush)

590+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"go"},"finish_reason":null}]}

499− url: https://hyper.charm.land/v1/chat/completions

591+

500− method: POST

592+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"("},"finish_reason":null}]}

501− response:

593+

502− proto: HTTP/2.0

594+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"3"},"finish_reason":null}]}

503− proto_major: 2

595+

504− proto_minor: 0

596+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

505− content_length: -1

597+

506− uncompressed: true

598+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"16"},"finish_reason":null}]}

507− body: |+

599+

508− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

600+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

601+

602+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

603+

604+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

605+

606+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"17"},"finish_reason":null}]}

607+

608+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

609+

610+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

611+

612+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"files"},"finish_reason":null}]}

613+

614+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"("},"finish_reason":null}]}

615+

616+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"5"},"finish_reason":null}]}

617+

618+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

619+

620+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"22"},"finish_reason":null}]}

621+

622+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

623+

624+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

625+

626+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

627+

628+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"23"},"finish_reason":null}]}

629+

630+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

509

631

510− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_3953728ce08b426ead57d862","type":"function","index":0,"function":{"name":"glob","arguments":""}}]},"finish_reason":null}]}

632+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

511

633

512− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

634+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"in"},"finish_reason":null}]}

513

635

514− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

636+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"("},"finish_reason":null}]}

515

637

516− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"path\": "}}]},"finish_reason":null}]}

638+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"2"},"finish_reason":null}]}

517

639

518− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"/tmp/c"}}]},"finish_reason":null}]}

640+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

519

641

520− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"rush"}}]},"finish_reason":null}]}

642+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"25"},"finish_reason":null}]}

521

643

522− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-test"}}]},"finish_reason":null}]}

644+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

523

645

524− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/Test"}}]},"finish_reason":null}]}

646+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

525

647

526− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Coder"}}]},"finish_reason":null}]}

648+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

527

649

528− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Agent"}}]},"finish_reason":null}]}

650+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"26"},"finish_reason":null}]}

529

651

530− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/glm"}}]},"finish_reason":null}]}

652+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

531

653

532− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-5.1"}}]},"finish_reason":null}]}

654+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

533

655

534− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/glob_tool"}}]},"finish_reason":null}]}

656+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"current"},"finish_reason":null}]}

535

657

536− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\", "}}]},"finish_reason":null}]}

658+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"("},"finish_reason":null}]}

537

659

538− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"pattern\": "}}]},"finish_reason":null}]}

660+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"7"},"finish_reason":null}]}

539

661

540− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"*.go\""}}]},"finish_reason":null}]}

662+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

541

663

542− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"}"}}]},"finish_reason":null}]}

664+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"33"},"finish_reason":null}]}

543

665

544− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

666+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

545

667

668+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

669+

670+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

671+

672+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"34"},"finish_reason":null}]}

673+

674+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

675+

676+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

677+

678+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"directory"},"finish_reason":null}]}

679+

680+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"("},"finish_reason":null}]}

681+

682+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"9"},"finish_reason":null}]}

683+

684+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

685+

686+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"43"},"finish_reason":null}]}

687+

688+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

689+

690+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

691+

692+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

693+

694+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"43"},"finish_reason":null}]}

695+

696+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

697+

698+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

699+

700+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Good"},"finish_reason":null}]}

701+

702+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".\n\n"},"finish_reason":null}]}

703+

704+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Return"},"finish_reason":null}]}

705+

706+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" just"},"finish_reason":null}]}

707+

708+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" that"},"finish_reason":null}]}

709+

710+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

711+

712+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

713+

714+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Glob"},"finish_reason":null}]}

715+

716+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" to"},"finish_reason":null}]}

717+

718+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" find"},"finish_reason":null}]}

719+

720+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" ."},"finish_reason":null}]}

721+

722+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"go"},"finish_reason":null}]}

723+

724+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" files"},"finish_reason":null}]}

725+

726+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" in"},"finish_reason":null}]}

727+

728+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" current"},"finish_reason":null}]}

729+

730+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" directory"},"finish_reason":null}]}

731+

732+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

733+

546− data: {"id":"chatcmpl-6d49471923fc1e2f062b1e54","object":"chat.completion.chunk","created":1779218888,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":713,"completion_tokens":33,"total_tokens":7850,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":32},"cost":{"usd":0.0013555,"hypercredits":0.02711},"remaining":{"hypercredits":0}}}

734+ data: {"id":"chatcmpl-72988fc242d434382e1cef89","object":"chat.completion.chunk","created":1780069690,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":216,"completion_tokens":315,"total_tokens":531,"completion_tokens_details":{"reasoning_tokens":297},"cost":{"usd":0.0004122,"hypercredits":0.008244},"remaining":{"hypercredits":0}}}

547

735

548 data: [DONE]

736 data: [DONE]

549

737

@@ -552,15 +740,15 @@ interactions:

552 - text/event-stream

740 - text/event-stream

553 status: 200 OK

741 status: 200 OK

554 code: 200

742 code: 200

555− duration: 2.891766166s

743+ duration: 428.407833ms

556- id: 2

744- id: 2

557 request:

745 request:

558 proto: HTTP/1.1

746 proto: HTTP/1.1

559 proto_major: 1

747 proto_major: 1

560 proto_minor: 1

748 proto_minor: 1

561− content_length: 34319

749+ content_length: 34347

562 host: ""

750 host: ""

563− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/glob_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use glob to find all .go files in the current directory\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_3953728ce08b426ead57d862\",\"function\":{\"arguments\":\"{\\\"path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/glob_tool\\\", \\\"pattern\\\": \\\"*.go\\\"}\",\"name\":\"glob\"},\"type\":\"function\"}],\"role\":\"assistant\"},{\"content\":\"/tmp/crush-test/TestCoderAgent/glm-5.1/glob_tool/main.go\",\"tool_call_id\":\"call_3953728ce08b426ead57d862\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

751+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/glob_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use glob to find all .go files in the current directory\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_fl8kSELeCV44nzdl56WTZKve\",\"function\":{\"arguments\":\"{\\\"pattern\\\": \\\"**/*.go\\\"}\",\"name\":\"glob\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"The user wants me to find all .go files in the current directory using the glob tool.\"},{\"content\":\"/tmp/crush-test/TestCoderAgent/glm-5.1/glob_tool/main.go\",\"tool_call_id\":\"call_fl8kSELeCV44nzdl56WTZKve\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

564 headers:

752 headers:

565 Accept:

753 Accept:

566 - application/json

754 - application/json

@@ -577,25 +765,35 @@ interactions:

577 content_length: -1

765 content_length: -1

578 uncompressed: true

766 uncompressed: true

579 body: |+

767 body: |+

580− data: {"id":"chatcmpl-69af89e5811d220aae3a87f0","object":"chat.completion.chunk","created":1779218891,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

768+ data: {"id":"chatcmpl-6fcea68a19fb7f5d4edd2860","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

769+

770+ data: {"id":"chatcmpl-6fcea68a19fb7f5d4edd2860","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Found"},"finish_reason":null}]}

771+

772+ data: {"id":"chatcmpl-6fcea68a19fb7f5d4edd2860","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" "},"finish_reason":null}]}

773+

774+ data: {"id":"chatcmpl-6fcea68a19fb7f5d4edd2860","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"1"},"finish_reason":null}]}

775+

776+ data: {"id":"chatcmpl-6fcea68a19fb7f5d4edd2860","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `."},"finish_reason":null}]}

777+

778+ data: {"id":"chatcmpl-6fcea68a19fb7f5d4edd2860","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"go"},"finish_reason":null}]}

581

779

582− data: {"id":"chatcmpl-69af89e5811d220aae3a87f0","object":"chat.completion.chunk","created":1779218891,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Found"},"finish_reason":null}]}

780+ data: {"id":"chatcmpl-6fcea68a19fb7f5d4edd2860","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`"},"finish_reason":null}]}

583

781

584− data: {"id":"chatcmpl-69af89e5811d220aae3a87f0","object":"chat.completion.chunk","created":1779218891,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" 1"},"finish_reason":null}]}

782+ data: {"id":"chatcmpl-6fcea68a19fb7f5d4edd2860","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" file"},"finish_reason":null}]}

585

783

586− data: {"id":"chatcmpl-69af89e5811d220aae3a87f0","object":"chat.completion.chunk","created":1779218891,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `.go"},"finish_reason":null}]}

784+ data: {"id":"chatcmpl-6fcea68a19fb7f5d4edd2860","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":":"},"finish_reason":null}]}

587

785

588− data: {"id":"chatcmpl-69af89e5811d220aae3a87f0","object":"chat.completion.chunk","created":1779218891,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"` file"},"finish_reason":null}]}

786+ data: {"id":"chatcmpl-6fcea68a19fb7f5d4edd2860","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

589

787

590− data: {"id":"chatcmpl-69af89e5811d220aae3a87f0","object":"chat.completion.chunk","created":1779218891,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":": `"},"finish_reason":null}]}

788+ data: {"id":"chatcmpl-6fcea68a19fb7f5d4edd2860","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"main"},"finish_reason":null}]}

591

789

592− data: {"id":"chatcmpl-69af89e5811d220aae3a87f0","object":"chat.completion.chunk","created":1779218891,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"main.go"},"finish_reason":null}]}

790+ data: {"id":"chatcmpl-6fcea68a19fb7f5d4edd2860","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".go"},"finish_reason":null}]}

593

791

594− data: {"id":"chatcmpl-69af89e5811d220aae3a87f0","object":"chat.completion.chunk","created":1779218891,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`"},"finish_reason":null}]}

792+ data: {"id":"chatcmpl-6fcea68a19fb7f5d4edd2860","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`"},"finish_reason":null}]}

595

793

596− data: {"id":"chatcmpl-69af89e5811d220aae3a87f0","object":"chat.completion.chunk","created":1779218891,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

794+ data: {"id":"chatcmpl-6fcea68a19fb7f5d4edd2860","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

597

795

598− data: {"id":"chatcmpl-69af89e5811d220aae3a87f0","object":"chat.completion.chunk","created":1779218891,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":64,"completion_tokens":14,"total_tokens":7886,"prompt_tokens_details":{"cached_tokens":7808},"completion_tokens_details":{"reasoning_tokens":13},"cost":{"usd":0.00021480000000000002,"hypercredits":0.004296},"remaining":{"hypercredits":0}}}

796+ data: {"id":"chatcmpl-6fcea68a19fb7f5d4edd2860","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":39,"completion_tokens":14,"total_tokens":7861,"prompt_tokens_details":{"cached_tokens":7808},"cost":{"usd":0.00010989,"hypercredits":0.0021978},"remaining":{"hypercredits":0}}}

599

797

600 data: [DONE]

798 data: [DONE]

601

799

@@ -604,4 +802,4 @@ interactions:

604 - text/event-stream

802 - text/event-stream

605 status: 200 OK

803 status: 200 OK

606 code: 200

804 code: 200

607− duration: 2.299728583s

805+ duration: 619.939042ms

internal/agent/testdata/TestCoderAgent/glm-5.1/grep_tool.yaml+182−138

@@ -6,9 +6,9 @@ interactions:

6 proto: HTTP/1.1

7 proto_major: 1

8 proto_minor: 1

9− content_length: 812

9+ content_length: 33944

10 host: ""

11− body: '{"messages":[{"content":"You will generate a short title based on the first message a user begins a conversation with.\n\n<rules>\n- Keep the title in the same language that the user wrote their message in.\n- Ensure it is not more than 50 characters long.\n- The title should be a summary of the user''s message.\n- It should be one line long.\n- Do not use quotes or colons.\n- The entire text you return will be used as the title.\n- Never return anything that is more than one sentence (one line) long.\n</rules>\n\n /no_think","role":"system"},{"content":"Generate a concise title for the following content:\n\nuse grep to search for the word ''package'' in go files\n <think>\n\n</think>","role":"user"}],"model":"gpt-oss-120b","max_completion_tokens":40,"stream_options":{"include_usage":true},"stream":true}'

11+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/grep_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use grep to search for the word 'package' in go files\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

12 headers:

13 Accept:

14 - application/json

@@ -25,217 +25,245 @@ interactions:

25 content_length: -1

26 uncompressed: true

27 body: |+

28− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

28+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

30− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"User"},"finish_reason":null}]}

30+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

32− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" asks"},"finish_reason":null}]}

32+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

34− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

34+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" wants"},"finish_reason":null}]}

36− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

36+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

38− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Generate"},"finish_reason":null}]}

38+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" search"},"finish_reason":null}]}

40− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

40+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

42− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

42+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

44− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

44+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" word"},"finish_reason":null}]}

46− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

46+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" '"},"finish_reason":null}]}

48− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

48+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"package"},"finish_reason":null}]}

50− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" following"},"finish_reason":null}]}

50+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"'"},"finish_reason":null}]}

52− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

52+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

54− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

54+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" Go"},"finish_reason":null}]}

56− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" use"},"finish_reason":null}]}

56+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

58− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" grep"},"finish_reason":null}]}

58+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" using"},"finish_reason":null}]}

60− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

60+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

62− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" search"},"finish_reason":null}]}

62+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" grep"},"finish_reason":null}]}

64− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

64+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" tool"},"finish_reason":null}]}

66− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

66+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

68− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" word"},"finish_reason":null}]}

68+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_21rwHdFtgIHBF2shcwEBHHd0","type":"function","index":0,"function":{"name":"grep","arguments":""}}]},"finish_reason":null}]}

70− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" '"},"finish_reason":null}]}

70+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

72− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"package"},"finish_reason":null}]}

72+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

74− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"'"},"finish_reason":null}]}

74+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"pattern\": "}}]},"finish_reason":null}]}

76− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

76+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"package"}}]},"finish_reason":null}]}

78− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" go"},"finish_reason":null}]}

78+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\", \"include\": "}}]},"finish_reason":null}]}

80− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

80+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"*."}}]},"finish_reason":null}]}

82− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

82+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"go"}}]},"finish_reason":null}]}

84− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" So"},"finish_reason":null}]}

84+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"}"}}]},"finish_reason":null}]}

86− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

86+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

88− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" should"},"finish_reason":null}]}

88+ data: {"id":"chatcmpl-0798a581811662a07d1abc32","object":"chat.completion.chunk","created":1780069691,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":709,"completion_tokens":37,"total_tokens":7850,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":20},"cost":{"usd":0.0011681900000000002,"hypercredits":0.0233638},"remaining":{"hypercredits":0}}}

90− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

90+ data: [DONE]

92+ headers:

93+ Content-Type:

94+ - text/event-stream

95+ status: 200 OK

96+ code: 200

97+ duration: 530.110208ms

98+- id: 1

99+ request:

100+ proto: HTTP/1.1

101+ proto_major: 1

102+ proto_minor: 1

103+ content_length: 812

104+ host: ""

105+ body: '{"messages":[{"content":"You will generate a short title based on the first message a user begins a conversation with.\n\n<rules>\n- Keep the title in the same language that the user wrote their message in.\n- Ensure it is not more than 50 characters long.\n- The title should be a summary of the user''s message.\n- It should be one line long.\n- Do not use quotes or colons.\n- The entire text you return will be used as the title.\n- Never return anything that is more than one sentence (one line) long.\n</rules>\n\n /no_think","role":"system"},{"content":"Generate a concise title for the following content:\n\nuse grep to search for the word ''package'' in go files\n <think>\n\n</think>","role":"user"}],"model":"gpt-oss-120b","max_completion_tokens":40,"stream_options":{"include_usage":true},"stream":true}'

106+ headers:

107+ Accept:

108+ - application/json

109+ Content-Type:

110+ - application/json

111+ User-Agent:

112+ - Charm-Crush/devel (https://charm.land/crush)

113+ url: https://hyper.charm.land/v1/chat/completions

114+ method: POST

115+ response:

116+ proto: HTTP/2.0

117+ proto_major: 2

118+ proto_minor: 0

119+ content_length: -1

120+ uncompressed: true

121+ body: |+

92− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" short"},"finish_reason":null}]}

122+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

123

94− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

124+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"User"},"finish_reason":null}]}

125

96− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \u003c"},"finish_reason":null}]}

126+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" wants"},"finish_reason":null}]}

127

98− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

128+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

129

100− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

130+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

101

131

102− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

132+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

103

133

104− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" summarize"},"finish_reason":null}]}

134+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

105

135

106− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

136+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

107

137

108− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Maybe"},"finish_reason":null}]}

138+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

109

139

110− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

140+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

111

141

112− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

142+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

113

143

114− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"G"},"finish_reason":null}]}

144+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"use"},"finish_reason":null}]}

115

145

116− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"rep"},"finish_reason":null}]}

146+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" grep"},"finish_reason":null}]}

117

147

118− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

148+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

119

149

120− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" '"},"finish_reason":null}]}

150+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" search"},"finish_reason":null}]}

121

151

122− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"package"},"finish_reason":null}]}

152+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

123

153

124− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"'"},"finish_reason":null}]}

154+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

125

155

126− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

156+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" word"},"finish_reason":null}]}

127

157

128− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Go"},"finish_reason":null}]}

158+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" '"},"finish_reason":null}]}

129

159

130− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" source"},"finish_reason":null}]}

160+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"package"},"finish_reason":null}]}

131

161

132− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

162+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"'"},"finish_reason":null}]}

133

163

134− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

164+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

135

165

136− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

166+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" go"},"finish_reason":null}]}

137

167

138− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

168+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

139

169

140− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"33"},"finish_reason":null}]}

170+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

141

171

142− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

172+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Must"},"finish_reason":null}]}

143

173

144− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

174+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

145

175

146− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" No"},"finish_reason":null}]}

176+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" under"},"finish_reason":null}]}

147

177

148− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

178+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

149

179

150− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" or"},"finish_reason":null}]}

180+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

151

181

152− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" col"},"finish_reason":null}]}

182+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

153

183

154− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ons"},"finish_reason":null}]}

184+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

155

185

156− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

186+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" one"},"finish_reason":null}]}

157

187

158− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Good"},"finish_reason":null}]}

188+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

159

189

160− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

190+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

161

191

162− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"G"},"finish_reason":null}]}

192+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

163

193

164− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"rep"},"finish_reason":null}]}

194+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

165

195

166− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" for"},"finish_reason":null}]}

196+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

167

197

168− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" package"},"finish_reason":null}]}

198+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

169

199

170− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" in"},"finish_reason":null}]}

200+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" colon"},"finish_reason":null}]}

171

201

172− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" Go"},"finish_reason":null}]}

202+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

173

203

174− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" source"},"finish_reason":null}]}

204+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Title"},"finish_reason":null}]}

175

205

176− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" files"},"finish_reason":null}]}

206+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" summar"},"finish_reason":null}]}

177

207

178− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

208+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"izing"},"finish_reason":null}]}

179

209

180− data: {"id":"chatcmpl-f81415e67853cf5721fdf330","object":"chat.completion.chunk","created":1779218892,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":216,"completion_tokens":84,"total_tokens":300,"completion_tokens_details":{"reasoning_tokens":66},"cost":{"usd":0.0001,"hypercredits":0.002},"remaining":{"hypercredits":0}}}

210+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

181

211

182− data: [DONE]

212+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

183

213

184− headers:

214+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"G"},"finish_reason":null}]}

185− Content-Type:

215+

186− - text/event-stream

216+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"rep"},"finish_reason":null}]}

187− status: 200 OK

217+

188− code: 200

218+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

189− duration: 662.429292ms

190−- id: 1

191− request:

192− proto: HTTP/1.1

193− proto_major: 1

194− proto_minor: 1

195− content_length: 33963

196− host: ""

197− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/grep_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use grep to search for the word 'package' in go files\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

198− headers:

199− Accept:

200− - application/json

201− Content-Type:

202− - application/json

203− User-Agent:

204− - Charm-Crush/devel (https://charm.land/crush)

205− url: https://hyper.charm.land/v1/chat/completions

206− method: POST

207− response:

208− proto: HTTP/2.0

209− proto_major: 2

210− proto_minor: 0

211− content_length: -1

212− uncompressed: true

213− body: |+

214− data: {"id":"chatcmpl-2e94aef325f8c1f92b99f5a8","object":"chat.completion.chunk","created":1779218894,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

215

219

216− data: {"id":"chatcmpl-2e94aef325f8c1f92b99f5a8","object":"chat.completion.chunk","created":1779218894,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_735d81ad9ab84c8a8477c7dc","type":"function","index":0,"function":{"name":"grep","arguments":""}}]},"finish_reason":null}]}

220+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" '"},"finish_reason":null}]}

217

221

218− data: {"id":"chatcmpl-2e94aef325f8c1f92b99f5a8","object":"chat.completion.chunk","created":1779218894,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

222+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"package"},"finish_reason":null}]}

219

223

220− data: {"id":"chatcmpl-2e94aef325f8c1f92b99f5a8","object":"chat.completion.chunk","created":1779218894,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

224+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"'"},"finish_reason":null}]}

221

225

222− data: {"id":"chatcmpl-2e94aef325f8c1f92b99f5a8","object":"chat.completion.chunk","created":1779218894,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"include\": "}}]},"finish_reason":null}]}

226+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

223

227

224− data: {"id":"chatcmpl-2e94aef325f8c1f92b99f5a8","object":"chat.completion.chunk","created":1779218894,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"*."}}]},"finish_reason":null}]}

228+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Go"},"finish_reason":null}]}

225

229

226− data: {"id":"chatcmpl-2e94aef325f8c1f92b99f5a8","object":"chat.completion.chunk","created":1779218894,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"go\""}}]},"finish_reason":null}]}

230+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

227

231

228− data: {"id":"chatcmpl-2e94aef325f8c1f92b99f5a8","object":"chat.completion.chunk","created":1779218894,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":", \"pattern\": "}}]},"finish_reason":null}]}

232+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

229

233

230− data: {"id":"chatcmpl-2e94aef325f8c1f92b99f5a8","object":"chat.completion.chunk","created":1779218894,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"package"}}]},"finish_reason":null}]}

234+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

231

235

232− data: {"id":"chatcmpl-2e94aef325f8c1f92b99f5a8","object":"chat.completion.chunk","created":1779218894,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\""}}]},"finish_reason":null}]}

236+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" under"},"finish_reason":null}]}

233

237

234− data: {"id":"chatcmpl-2e94aef325f8c1f92b99f5a8","object":"chat.completion.chunk","created":1779218894,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"}"}}]},"finish_reason":null}]}

238+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

235

239

236− data: {"id":"chatcmpl-2e94aef325f8c1f92b99f5a8","object":"chat.completion.chunk","created":1779218894,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

240+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

237

241

242+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

243+

244+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Return"},"finish_reason":null}]}

245+

246+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" that"},"finish_reason":null}]}

247+

248+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

249+

250+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"G"},"finish_reason":null}]}

251+

252+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"rep"},"finish_reason":null}]}

253+

254+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" for"},"finish_reason":null}]}

255+

256+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" package"},"finish_reason":null}]}

257+

258+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" in"},"finish_reason":null}]}

259+

260+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" Go"},"finish_reason":null}]}

261+

262+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" files"},"finish_reason":null}]}

263+

264+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

265+

238− data: {"id":"chatcmpl-2e94aef325f8c1f92b99f5a8","object":"chat.completion.chunk","created":1779218894,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":714,"completion_tokens":18,"total_tokens":7836,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":17},"cost":{"usd":0.001225,"hypercredits":0.024499999999999997},"remaining":{"hypercredits":0}}}

266+ data: {"id":"chatcmpl-d9dfa682ad16493bb43839d1","object":"chat.completion.chunk","created":1780069692,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":217,"completion_tokens":80,"total_tokens":297,"completion_tokens_details":{"reasoning_tokens":64},"cost":{"usd":0.0001269,"hypercredits":0.002538},"remaining":{"hypercredits":0}}}

239

267

240 data: [DONE]

268 data: [DONE]

241

269

@@ -244,15 +272,15 @@ interactions:

244 - text/event-stream

272 - text/event-stream

245 status: 200 OK

273 status: 200 OK

246 code: 200

274 code: 200

247− duration: 2.470623166s

275+ duration: 910.652042ms

248- id: 2

276- id: 2

249 request:

277 request:

250 proto: HTTP/1.1

278 proto: HTTP/1.1

251 proto_major: 1

279 proto_major: 1

252 proto_minor: 1

280 proto_minor: 1

253− content_length: 34331

281+ content_length: 34415

254 host: ""

282 host: ""

255− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/grep_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use grep to search for the word 'package' in go files\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_735d81ad9ab84c8a8477c7dc\",\"function\":{\"arguments\":\"{\\\"include\\\": \\\"*.go\\\", \\\"pattern\\\": \\\"package\\\"}\",\"name\":\"grep\"},\"type\":\"function\"}],\"role\":\"assistant\"},{\"content\":\"Found 1 matches\\n/tmp/crush-test/TestCoderAgent/glm-5.1/grep_tool/main.go:\\n Line 1, Char 1: package main\\n\",\"tool_call_id\":\"call_735d81ad9ab84c8a8477c7dc\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

283+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/grep_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use grep to search for the word 'package' in go files\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_21rwHdFtgIHBF2shcwEBHHd0\",\"function\":{\"arguments\":\"{\\\"pattern\\\": \\\"package\\\", \\\"include\\\": \\\"*.go\\\"}\",\"name\":\"grep\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"The user wants to search for the word 'package' in Go files using the grep tool.\"},{\"content\":\"Found 1 matches\\n/tmp/crush-test/TestCoderAgent/glm-5.1/grep_tool/main.go:\\n Line 1, Char 1: package main\\n\",\"tool_call_id\":\"call_21rwHdFtgIHBF2shcwEBHHd0\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

256 headers:

284 headers:

257 Accept:

285 Accept:

258 - application/json

286 - application/json

@@ -269,27 +297,43 @@ interactions:

269 content_length: -1

297 content_length: -1

270 uncompressed: true

298 uncompressed: true

271 body: |+

299 body: |+

272− data: {"id":"chatcmpl-911d6361c2318f054897ddfb","object":"chat.completion.chunk","created":1779218898,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

300+ data: {"id":"chatcmpl-14835ce4b37a296037351a3c","object":"chat.completion.chunk","created":1780069692,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

301+

302+ data: {"id":"chatcmpl-14835ce4b37a296037351a3c","object":"chat.completion.chunk","created":1780069692,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Found"},"finish_reason":null}]}

303+

304+ data: {"id":"chatcmpl-14835ce4b37a296037351a3c","object":"chat.completion.chunk","created":1780069692,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" "},"finish_reason":null}]}

305+

306+ data: {"id":"chatcmpl-14835ce4b37a296037351a3c","object":"chat.completion.chunk","created":1780069692,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"1"},"finish_reason":null}]}

307+

308+ data: {"id":"chatcmpl-14835ce4b37a296037351a3c","object":"chat.completion.chunk","created":1780069692,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" match"},"finish_reason":null}]}

309+

310+ data: {"id":"chatcmpl-14835ce4b37a296037351a3c","object":"chat.completion.chunk","created":1780069692,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":":"},"finish_reason":null}]}

311+

312+ data: {"id":"chatcmpl-14835ce4b37a296037351a3c","object":"chat.completion.chunk","created":1780069692,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

313+

314+ data: {"id":"chatcmpl-14835ce4b37a296037351a3c","object":"chat.completion.chunk","created":1780069692,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"main"},"finish_reason":null}]}

315+

316+ data: {"id":"chatcmpl-14835ce4b37a296037351a3c","object":"chat.completion.chunk","created":1780069692,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".go"},"finish_reason":null}]}

273

317

274− data: {"id":"chatcmpl-911d6361c2318f054897ddfb","object":"chat.completion.chunk","created":1779218898,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"1"},"finish_reason":null}]}

318+ data: {"id":"chatcmpl-14835ce4b37a296037351a3c","object":"chat.completion.chunk","created":1780069692,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":":"},"finish_reason":null}]}

275

319

276− data: {"id":"chatcmpl-911d6361c2318f054897ddfb","object":"chat.completion.chunk","created":1779218898,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" match found"},"finish_reason":null}]}

320+ data: {"id":"chatcmpl-14835ce4b37a296037351a3c","object":"chat.completion.chunk","created":1780069692,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"1"},"finish_reason":null}]}

277

321

278− data: {"id":"chatcmpl-911d6361c2318f054897ddfb","object":"chat.completion.chunk","created":1779218898,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":": `main"},"finish_reason":null}]}

322+ data: {"id":"chatcmpl-14835ce4b37a296037351a3c","object":"chat.completion.chunk","created":1780069692,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`"},"finish_reason":null}]}

279

323

280− data: {"id":"chatcmpl-911d6361c2318f054897ddfb","object":"chat.completion.chunk","created":1779218898,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".go"},"finish_reason":null}]}

324+ data: {"id":"chatcmpl-14835ce4b37a296037351a3c","object":"chat.completion.chunk","created":1780069692,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" —"},"finish_reason":null}]}

281

325

282− data: {"id":"chatcmpl-911d6361c2318f054897ddfb","object":"chat.completion.chunk","created":1779218898,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":":"},"finish_reason":null}]}

326+ data: {"id":"chatcmpl-14835ce4b37a296037351a3c","object":"chat.completion.chunk","created":1780069692,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

283

327

284− data: {"id":"chatcmpl-911d6361c2318f054897ddfb","object":"chat.completion.chunk","created":1779218898,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"1"},"finish_reason":null}]}

328+ data: {"id":"chatcmpl-14835ce4b37a296037351a3c","object":"chat.completion.chunk","created":1780069692,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"package"},"finish_reason":null}]}

285

329

286− data: {"id":"chatcmpl-911d6361c2318f054897ddfb","object":"chat.completion.chunk","created":1779218898,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"` — `package"},"finish_reason":null}]}

330+ data: {"id":"chatcmpl-14835ce4b37a296037351a3c","object":"chat.completion.chunk","created":1780069692,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" main"},"finish_reason":null}]}

287

331

288− data: {"id":"chatcmpl-911d6361c2318f054897ddfb","object":"chat.completion.chunk","created":1779218898,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" main`"},"finish_reason":null}]}

332+ data: {"id":"chatcmpl-14835ce4b37a296037351a3c","object":"chat.completion.chunk","created":1780069692,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`"},"finish_reason":null}]}

289

333

290− data: {"id":"chatcmpl-911d6361c2318f054897ddfb","object":"chat.completion.chunk","created":1779218898,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

334+ data: {"id":"chatcmpl-14835ce4b37a296037351a3c","object":"chat.completion.chunk","created":1780069692,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

291

335

292− data: {"id":"chatcmpl-911d6361c2318f054897ddfb","object":"chat.completion.chunk","created":1779218898,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":68,"completion_tokens":17,"total_tokens":7893,"prompt_tokens_details":{"cached_tokens":7808},"completion_tokens_details":{"reasoning_tokens":16},"cost":{"usd":0.00024720000000000005,"hypercredits":0.0049440000000000005},"remaining":{"hypercredits":0}}}

336+ data: {"id":"chatcmpl-14835ce4b37a296037351a3c","object":"chat.completion.chunk","created":1780069692,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":63,"completion_tokens":18,"total_tokens":7889,"prompt_tokens_details":{"cached_tokens":7808},"cost":{"usd":0.00015813,"hypercredits":0.0031626},"remaining":{"hypercredits":0}}}

293

337

294 data: [DONE]

338 data: [DONE]

295

339

@@ -298,4 +342,4 @@ interactions:

298 - text/event-stream

342 - text/event-stream

299 status: 200 OK

343 status: 200 OK

300 code: 200

344 code: 200

301− duration: 3.423796667s

345+ duration: 525.813667ms

internal/agent/testdata/TestCoderAgent/glm-5.1/ls_tool.yaml+271−101

@@ -6,9 +6,9 @@ interactions:

6 proto: HTTP/1.1

7 proto_major: 1

8 proto_minor: 1

9− content_length: 808

9+ content_length: 33938

10 host: ""

11− body: '{"messages":[{"content":"You will generate a short title based on the first message a user begins a conversation with.\n\n<rules>\n- Keep the title in the same language that the user wrote their message in.\n- Ensure it is not more than 50 characters long.\n- The title should be a summary of the user''s message.\n- It should be one line long.\n- Do not use quotes or colons.\n- The entire text you return will be used as the title.\n- Never return anything that is more than one sentence (one line) long.\n</rules>\n\n /no_think","role":"system"},{"content":"Generate a concise title for the following content:\n\nuse ls to list the files in the current directory\n <think>\n\n</think>","role":"user"}],"model":"gpt-oss-120b","max_completion_tokens":40,"stream_options":{"include_usage":true},"stream":true}'

11+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/ls_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use ls to list the files in the current directory\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

12 headers:

13 Accept:

14 - application/json

@@ -25,117 +25,89 @@ interactions:

25 content_length: -1

26 uncompressed: true

27 body: |+

28− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

28+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

29−

30− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

31−

32− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

33−

34− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" wants"},"finish_reason":null}]}

35−

36− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

37−

38− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

39−

40− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

41−

42− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

43−

44− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" that"},"finish_reason":null}]}

45−

46− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

47−

48− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

49−

50− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Must"},"finish_reason":null}]}

51−

52− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

53−

54− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" under"},"finish_reason":null}]}

56− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

30+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

58− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

32+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

60− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

34+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" wants"},"finish_reason":null}]}

62− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

36+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" me"},"finish_reason":null}]}

64− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" one"},"finish_reason":null}]}

38+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

66− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

40+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" list"},"finish_reason":null}]}

68− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

42+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

70− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

44+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

72− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

46+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

74− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

48+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" current"},"finish_reason":null}]}

76− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Summ"},"finish_reason":null}]}

50+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" directory"},"finish_reason":null}]}

78− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ar"},"finish_reason":null}]}

52+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" using"},"finish_reason":null}]}

80− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ize"},"finish_reason":null}]}

54+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

82− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

56+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" ls"},"finish_reason":null}]}

84− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

58+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" tool"},"finish_reason":null}]}

86− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Using"},"finish_reason":null}]}

60+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

88− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ls"},"finish_reason":null}]}

62+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_DolI6GDGB8olX8skQ056EtPD","type":"function","index":0,"function":{"name":"ls","arguments":""}}]},"finish_reason":null}]}

90− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

64+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

92− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" list"},"finish_reason":null}]}

66+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

94− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

68+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"depth\": "}}]},"finish_reason":null}]}

96− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

70+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"1"}}]},"finish_reason":null}]}

98− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" current"},"finish_reason":null}]}

72+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":", \"path\": "}}]},"finish_reason":null}]}

100− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" directory"},"finish_reason":null}]}

74+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"/tmp"}}]},"finish_reason":null}]}

101

102− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

76+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/c"}}]},"finish_reason":null}]}

103

104− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That"},"finish_reason":null}]}

78+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"rush"}}]},"finish_reason":null}]}

105

106− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" is"},"finish_reason":null}]}

80+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-test"}}]},"finish_reason":null}]}

107

108− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" within"},"finish_reason":null}]}

82+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/Test"}}]},"finish_reason":null}]}

109

110− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" limit"},"finish_reason":null}]}

84+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Coder"}}]},"finish_reason":null}]}

111

112− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".\n\n"},"finish_reason":null}]}

86+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Agent"}}]},"finish_reason":null}]}

113

114− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Return"},"finish_reason":null}]}

88+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/g"}}]},"finish_reason":null}]}

115

116− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" that"},"finish_reason":null}]}

90+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"lm"}}]},"finish_reason":null}]}

117

118− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

92+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-"}}]},"finish_reason":null}]}

119

120− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Using"},"finish_reason":null}]}

94+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"5"}}]},"finish_reason":null}]}

121

122− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" ls"},"finish_reason":null}]}

96+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"."}}]},"finish_reason":null}]}

123

124− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" to"},"finish_reason":null}]}

98+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"1"}}]},"finish_reason":null}]}

125

126− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" list"},"finish_reason":null}]}

100+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/"}}]},"finish_reason":null}]}

127

101

128− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" files"},"finish_reason":null}]}

102+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"ls"}}]},"finish_reason":null}]}

129

103

130− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" in"},"finish_reason":null}]}

104+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_tool"}}]},"finish_reason":null}]}

131

105

132− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" current"},"finish_reason":null}]}

106+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"}"}}]},"finish_reason":null}]}

133

107

134− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" directory"},"finish_reason":null}]}

108+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

135

109

136− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

137−

138− data: {"id":"chatcmpl-ffa5831af4f19760a22ffa9b","object":"chat.completion.chunk","created":1779218899,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":152,"completion_tokens":8,"total_tokens":160,"completion_tokens_details":{"reasoning_tokens":45},"cost":{"usd":0.0001,"hypercredits":0.002},"remaining":{"hypercredits":0}}}

110+ data: {"id":"chatcmpl-164a30f9461a73f5ee098b9c","object":"chat.completion.chunk","created":1780069693,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":706,"completion_tokens":48,"total_tokens":7858,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":19},"cost":{"usd":0.0012062600000000002,"hypercredits":0.024125200000000003},"remaining":{"hypercredits":0}}}

139

111

140 data: [DONE]

112 data: [DONE]

141

113

@@ -144,15 +116,15 @@ interactions:

144 - text/event-stream

116 - text/event-stream

145 status: 200 OK

117 status: 200 OK

146 code: 200

118 code: 200

147− duration: 703.770958ms

119+ duration: 809.482333ms

148- id: 1

120- id: 1

149 request:

121 request:

150 proto: HTTP/1.1

122 proto: HTTP/1.1

151 proto_major: 1

123 proto_major: 1

152 proto_minor: 1

124 proto_minor: 1

153− content_length: 33957

125+ content_length: 34399

154 host: ""

126 host: ""

155− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/ls_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use ls to list the files in the current directory\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

127+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/ls_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use ls to list the files in the current directory\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_DolI6GDGB8olX8skQ056EtPD\",\"function\":{\"arguments\":\"{\\\"depth\\\": 1, \\\"path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/ls_tool\\\"}\",\"name\":\"ls\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"The user wants me to list files in the current directory using the ls tool.\"},{\"content\":\"\\n- /tmp/crush-test/TestCoderAgent/glm-5.1/ls_tool/\\n - go.mod\\n - main.go\\n\",\"tool_call_id\":\"call_DolI6GDGB8olX8skQ056EtPD\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

156 headers:

128 headers:

157 Accept:

129 Accept:

158 - application/json

130 - application/json

@@ -169,29 +141,31 @@ interactions:

169 content_length: -1

141 content_length: -1

170 uncompressed: true

142 uncompressed: true

171 body: |+

143 body: |+

172− data: {"id":"chatcmpl-53a07c6cf5196536e2ce3c26","object":"chat.completion.chunk","created":1779218900,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

144+ data: {"id":"chatcmpl-2c111a808431f0ffbafc2556","object":"chat.completion.chunk","created":1780069694,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

145+

146+ data: {"id":"chatcmpl-2c111a808431f0ffbafc2556","object":"chat.completion.chunk","created":1780069694,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"-"},"finish_reason":null}]}

173

147

174− data: {"id":"chatcmpl-53a07c6cf5196536e2ce3c26","object":"chat.completion.chunk","created":1779218900,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"Simple"},"finish_reason":null}]}

148+ data: {"id":"chatcmpl-2c111a808431f0ffbafc2556","object":"chat.completion.chunk","created":1780069694,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

175

149

176− data: {"id":"chatcmpl-53a07c6cf5196536e2ce3c26","object":"chat.completion.chunk","created":1779218900,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" request"},"finish_reason":null}]}

150+ data: {"id":"chatcmpl-2c111a808431f0ffbafc2556","object":"chat.completion.chunk","created":1780069694,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"go"},"finish_reason":null}]}

177

151

178− data: {"id":"chatcmpl-53a07c6cf5196536e2ce3c26","object":"chat.completion.chunk","created":1779218900,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" - list"},"finish_reason":null}]}

152+ data: {"id":"chatcmpl-2c111a808431f0ffbafc2556","object":"chat.completion.chunk","created":1780069694,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".mod"},"finish_reason":null}]}

179

153

180− data: {"id":"chatcmpl-53a07c6cf5196536e2ce3c26","object":"chat.completion.chunk","created":1779218900,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

154+ data: {"id":"chatcmpl-2c111a808431f0ffbafc2556","object":"chat.completion.chunk","created":1780069694,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`\n"},"finish_reason":null}]}

181

155

182− data: {"id":"chatcmpl-53a07c6cf5196536e2ce3c26","object":"chat.completion.chunk","created":1779218900,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" in current"},"finish_reason":null}]}

156+ data: {"id":"chatcmpl-2c111a808431f0ffbafc2556","object":"chat.completion.chunk","created":1780069694,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"-"},"finish_reason":null}]}

183

157

184− data: {"id":"chatcmpl-53a07c6cf5196536e2ce3c26","object":"chat.completion.chunk","created":1779218900,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" directory."},"finish_reason":null}]}

158+ data: {"id":"chatcmpl-2c111a808431f0ffbafc2556","object":"chat.completion.chunk","created":1780069694,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

185

159

186− data: {"id":"chatcmpl-53a07c6cf5196536e2ce3c26","object":"chat.completion.chunk","created":1779218900,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_0c7f7874c4db4882bf892ed9","type":"function","index":0,"function":{"name":"ls","arguments":""}}]},"finish_reason":null}]}

160+ data: {"id":"chatcmpl-2c111a808431f0ffbafc2556","object":"chat.completion.chunk","created":1780069694,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"main"},"finish_reason":null}]}

187

161

188− data: {"id":"chatcmpl-53a07c6cf5196536e2ce3c26","object":"chat.completion.chunk","created":1779218900,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

162+ data: {"id":"chatcmpl-2c111a808431f0ffbafc2556","object":"chat.completion.chunk","created":1780069694,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".go"},"finish_reason":null}]}

189

163

190− data: {"id":"chatcmpl-53a07c6cf5196536e2ce3c26","object":"chat.completion.chunk","created":1779218900,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{}"}}]},"finish_reason":null}]}

164+ data: {"id":"chatcmpl-2c111a808431f0ffbafc2556","object":"chat.completion.chunk","created":1780069694,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`"},"finish_reason":null}]}

191

165

192− data: {"id":"chatcmpl-53a07c6cf5196536e2ce3c26","object":"chat.completion.chunk","created":1779218900,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

166+ data: {"id":"chatcmpl-2c111a808431f0ffbafc2556","object":"chat.completion.chunk","created":1780069694,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

193

167

194− data: {"id":"chatcmpl-53a07c6cf5196536e2ce3c26","object":"chat.completion.chunk","created":1779218900,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":711,"completion_tokens":14,"total_tokens":7829,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":9},"cost":{"usd":0.0011677,"hypercredits":0.023354},"remaining":{"hypercredits":0}}}

168+ data: {"id":"chatcmpl-2c111a808431f0ffbafc2556","object":"chat.completion.chunk","created":1780069694,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":68,"completion_tokens":12,"total_tokens":7888,"prompt_tokens_details":{"cached_tokens":7808},"cost":{"usd":0.00013947999999999999,"hypercredits":0.0027895999999999997},"remaining":{"hypercredits":0}}}

195

169

196 data: [DONE]

170 data: [DONE]

197

171

@@ -200,15 +174,15 @@ interactions:

200 - text/event-stream

174 - text/event-stream

201 status: 200 OK

175 status: 200 OK

202 code: 200

176 code: 200

203− duration: 1.620500959s

177+ duration: 536.859958ms

204- id: 2

178- id: 2

205 request:

179 request:

206 proto: HTTP/1.1

180 proto: HTTP/1.1

207 proto_major: 1

181 proto_major: 1

208 proto_minor: 1

182 proto_minor: 1

209− content_length: 34318

183+ content_length: 808

210 host: ""

184 host: ""

211− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/ls_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use ls to list the files in the current directory\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_0c7f7874c4db4882bf892ed9\",\"function\":{\"arguments\":\"{}\",\"name\":\"ls\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"Simple request - list files in current directory.\"},{\"content\":\"\\n- /tmp/crush-test/TestCoderAgent/glm-5.1/ls_tool/\\n - go.mod\\n - main.go\\n\",\"tool_call_id\":\"call_0c7f7874c4db4882bf892ed9\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

185+ body: '{"messages":[{"content":"You will generate a short title based on the first message a user begins a conversation with.\n\n<rules>\n- Keep the title in the same language that the user wrote their message in.\n- Ensure it is not more than 50 characters long.\n- The title should be a summary of the user''s message.\n- It should be one line long.\n- Do not use quotes or colons.\n- The entire text you return will be used as the title.\n- Never return anything that is more than one sentence (one line) long.\n</rules>\n\n /no_think","role":"system"},{"content":"Generate a concise title for the following content:\n\nuse ls to list the files in the current directory\n <think>\n\n</think>","role":"user"}],"model":"gpt-oss-120b","max_completion_tokens":40,"stream_options":{"include_usage":true},"stream":true}'

212 headers:

186 headers:

213 Accept:

187 Accept:

214 - application/json

188 - application/json

@@ -225,25 +199,221 @@ interactions:

225 content_length: -1

199 content_length: -1

226 uncompressed: true

200 uncompressed: true

227 body: |+

201 body: |+

228− data: {"id":"chatcmpl-d3d746fe12a971be2cdbf977","object":"chat.completion.chunk","created":1779218903,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

202+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

203+

204+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

205+

206+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" system"},"finish_reason":null}]}

207+

208+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

209+

210+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" The"},"finish_reason":null}]}

211+

212+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" developer"},"finish_reason":null}]}

213+

214+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" says"},"finish_reason":null}]}

215+

216+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

217+

218+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" generate"},"finish_reason":null}]}

219+

220+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" short"},"finish_reason":null}]}

221+

222+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

223+

224+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" based"},"finish_reason":null}]}

225+

226+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" on"},"finish_reason":null}]}

227+

228+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" first"},"finish_reason":null}]}

229+

230+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" message"},"finish_reason":null}]}

231+

232+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

233+

234+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

235+

236+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" begins"},"finish_reason":null}]}

237+

238+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

239+

240+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" conversation"},"finish_reason":null}]}

241+

242+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

243+

244+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

245+

246+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" The"},"finish_reason":null}]}

247+

248+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

249+

250+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" says"},"finish_reason":null}]}

251+

252+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

253+

254+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

255+

256+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Generate"},"finish_reason":null}]}

257+

258+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

259+

260+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

261+

262+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

263+

264+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

265+

266+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

267+

268+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" following"},"finish_reason":null}]}

269+

270+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

271+

272+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

273+

274+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" use"},"finish_reason":null}]}

275+

276+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ls"},"finish_reason":null}]}

277+

278+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

279+

280+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" list"},"finish_reason":null}]}

281+

282+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

283+

284+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

285+

286+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

287+

288+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

289+

290+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" current"},"finish_reason":null}]}

291+

292+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" directory"},"finish_reason":null}]}

293+

294+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \u003c"},"finish_reason":null}]}

295+

296+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"think"},"finish_reason":null}]}

297+

298+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\u003e\""},"finish_reason":null}]}

299+

300+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" They"},"finish_reason":null}]}

301+

302+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" want"},"finish_reason":null}]}

303+

304+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

305+

306+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

307+

308+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

309+

310+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Should"},"finish_reason":null}]}

311+

312+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

313+

314+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

315+

316+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" same"},"finish_reason":null}]}

317+

318+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" language"},"finish_reason":null}]}

319+

320+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

321+

322+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"English"},"finish_reason":null}]}

323+

324+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")."},"finish_reason":null}]}

325+

326+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Not"},"finish_reason":null}]}

327+

328+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" more"},"finish_reason":null}]}

329+

330+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" than"},"finish_reason":null}]}

331+

332+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

333+

334+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

335+

336+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

337+

338+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

339+

340+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Should"},"finish_reason":null}]}

341+

342+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" summarize"},"finish_reason":null}]}

343+

344+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

345+

346+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" message"},"finish_reason":null}]}

347+

348+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

349+

350+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Title"},"finish_reason":null}]}

351+

352+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

353+

354+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

355+

356+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Using"},"finish_reason":null}]}

357+

358+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ls"},"finish_reason":null}]}

359+

360+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

361+

362+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" List"},"finish_reason":null}]}

363+

364+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Current"},"finish_reason":null}]}

365+

366+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Directory"},"finish_reason":null}]}

367+

368+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Files"},"finish_reason":null}]}

369+

370+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

371+

372+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

373+

374+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" under"},"finish_reason":null}]}

375+

376+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

377+

378+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

379+

380+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

381+

382+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

383+

384+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"38"},"finish_reason":null}]}

385+

386+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")."},"finish_reason":null}]}

387+

388+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" No"},"finish_reason":null}]}

389+

390+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

391+

392+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

393+

394+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" One"},"finish_reason":null}]}

395+

396+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

397+

398+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

229

399

230− data: {"id":"chatcmpl-d3d746fe12a971be2cdbf977","object":"chat.completion.chunk","created":1779218903,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Two"},"finish_reason":null}]}

400+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Using"},"finish_reason":null}]}

231

401

232− data: {"id":"chatcmpl-d3d746fe12a971be2cdbf977","object":"chat.completion.chunk","created":1779218903,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" files in"},"finish_reason":null}]}

402+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" ls"},"finish_reason":null}]}

233

403

234− data: {"id":"chatcmpl-d3d746fe12a971be2cdbf977","object":"chat.completion.chunk","created":1779218903,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" the current directory"},"finish_reason":null}]}

404+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" to"},"finish_reason":null}]}

235

405

236− data: {"id":"chatcmpl-d3d746fe12a971be2cdbf977","object":"chat.completion.chunk","created":1779218903,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":":\n- `go"},"finish_reason":null}]}

406+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" list"},"finish_reason":null}]}

237

407

238− data: {"id":"chatcmpl-d3d746fe12a971be2cdbf977","object":"chat.completion.chunk","created":1779218903,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".mod`\n-"},"finish_reason":null}]}

408+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" current"},"finish_reason":null}]}

239

409

240− data: {"id":"chatcmpl-d3d746fe12a971be2cdbf977","object":"chat.completion.chunk","created":1779218903,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `main"},"finish_reason":null}]}

410+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" directory"},"finish_reason":null}]}

241

411

242− data: {"id":"chatcmpl-d3d746fe12a971be2cdbf977","object":"chat.completion.chunk","created":1779218903,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".go`"},"finish_reason":null}]}

412+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" files"},"finish_reason":null}]}

243

413

244− data: {"id":"chatcmpl-d3d746fe12a971be2cdbf977","object":"chat.completion.chunk","created":1779218903,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

414+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

245

415

246− data: {"id":"chatcmpl-d3d746fe12a971be2cdbf977","object":"chat.completion.chunk","created":1779218903,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":46,"completion_tokens":19,"total_tokens":7873,"prompt_tokens_details":{"cached_tokens":7808},"completion_tokens_details":{"reasoning_tokens":18},"cost":{"usd":0.00023180000000000002,"hypercredits":0.004636},"remaining":{"hypercredits":0}}}

416+ data: {"id":"chatcmpl-b40bae129980d771e7d96e2c","object":"chat.completion.chunk","created":1780069695,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":214,"completion_tokens":115,"total_tokens":329,"completion_tokens_details":{"reasoning_tokens":99},"cost":{"usd":0.00016906,"hypercredits":0.0033812},"remaining":{"hypercredits":0}}}

247

417

248 data: [DONE]

418 data: [DONE]

249

419

@@ -252,4 +422,4 @@ interactions:

252 - text/event-stream

422 - text/event-stream

253 status: 200 OK

423 status: 200 OK

254 code: 200

424 code: 200

255− duration: 2.73167025s

425+ duration: 2.958506667s

internal/agent/testdata/TestCoderAgent/glm-5.1/multiedit_tool.yaml+675−363

@@ -2,6 +2,122 @@

2version: 2

3interactions:

4- id: 0

5+ request:

6+ proto: HTTP/1.1

7+ proto_major: 1

8+ proto_minor: 1

9+ content_length: 34024

10+ host: ""

11+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use multiedit to change 'Hello, World!' to 'Hello, Crush!' and add a comment '// Greeting' above the fmt.Println line in main.go\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

12+ headers:

13+ Accept:

14+ - application/json

15+ Content-Type:

16+ - application/json

17+ User-Agent:

18+ - Charm-Crush/devel (https://charm.land/crush)

19+ url: https://hyper.charm.land/v1/chat/completions

20+ method: POST

21+ response:

22+ proto: HTTP/2.0

23+ proto_major: 2

24+ proto_minor: 0

25+ content_length: -1

26+ uncompressed: true

27+ body: |+

28+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

29+

30+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"Let"},"finish_reason":null}]}

31+

32+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" me"},"finish_reason":null}]}

33+

34+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" first"},"finish_reason":null}]}

35+

36+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" read"},"finish_reason":null}]}

37+

38+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

39+

40+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

41+

42+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":".go"},"finish_reason":null}]}

43+

44+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" file"},"finish_reason":null}]}

45+

46+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

47+

48+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" understand"},"finish_reason":null}]}

49+

50+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" its"},"finish_reason":null}]}

51+

52+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" current"},"finish_reason":null}]}

53+

54+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

55+

56+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

57+

58+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_4BvAOulZ4vPbT3o9XRDa7mrF","type":"function","index":0,"function":{"name":"view","arguments":""}}]},"finish_reason":null}]}

59+

60+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

61+

62+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

63+

64+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"file_path\": "}}]},"finish_reason":null}]}

65+

66+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"/tmp"}}]},"finish_reason":null}]}

67+

68+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/c"}}]},"finish_reason":null}]}

69+

70+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"rush"}}]},"finish_reason":null}]}

71+

72+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-test"}}]},"finish_reason":null}]}

73+

74+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/Test"}}]},"finish_reason":null}]}

75+

76+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Coder"}}]},"finish_reason":null}]}

77+

78+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Agent"}}]},"finish_reason":null}]}

79+

80+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/g"}}]},"finish_reason":null}]}

81+

82+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"lm"}}]},"finish_reason":null}]}

83+

84+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-"}}]},"finish_reason":null}]}

85+

86+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"5"}}]},"finish_reason":null}]}

87+

88+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"."}}]},"finish_reason":null}]}

89+

90+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"1"}}]},"finish_reason":null}]}

91+

92+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/m"}}]},"finish_reason":null}]}

93+

94+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"ult"}}]},"finish_reason":null}]}

95+

96+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"ied"}}]},"finish_reason":null}]}

97+

98+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"it"}}]},"finish_reason":null}]}

99+

100+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_tool"}}]},"finish_reason":null}]}

101+

102+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/main"}}]},"finish_reason":null}]}

103+

104+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".go"}}]},"finish_reason":null}]}

105+

106+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"}"}}]},"finish_reason":null}]}

107+

108+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

109+

110+ data: {"id":"chatcmpl-e99200919451060d099e458c","object":"chat.completion.chunk","created":1780069697,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":731,"completion_tokens":45,"total_tokens":7880,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":18},"cost":{"usd":0.00122221,"hypercredits":0.0244442},"remaining":{"hypercredits":0}}}

111+

112+ data: [DONE]

113+

114+ headers:

115+ Content-Type:

116+ - text/event-stream

117+ status: 200 OK

118+ code: 200

119+ duration: 620.445917ms

120+- id: 1

5 request:

121 request:

6 proto: HTTP/1.1

122 proto: HTTP/1.1

7 proto_major: 1

123 proto_major: 1

@@ -25,505 +141,593 @@ interactions:

25 content_length: -1

141 content_length: -1

26 uncompressed: true

142 uncompressed: true

27 body: |+

143 body: |+

28− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

144+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

145

30− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

146+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"We"},"finish_reason":null}]}

147

32− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

148+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" need"},"finish_reason":null}]}

149

34− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" wants"},"finish_reason":null}]}

150+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

151

36− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

152+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" produce"},"finish_reason":null}]}

153

38− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

154+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

155

40− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

156+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

157

42− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

158+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

159

44− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

160+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" same"},"finish_reason":null}]}

161

46− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

162+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" language"},"finish_reason":null}]}

163

48− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

164+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

165

50− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

166+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"English"},"finish_reason":null}]}

167

52− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"use"},"finish_reason":null}]}

168+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")."},"finish_reason":null}]}

169

54− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" mult"},"finish_reason":null}]}

170+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Under"},"finish_reason":null}]}

171

56− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ied"},"finish_reason":null}]}

172+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

173

58− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"it"},"finish_reason":null}]}

174+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

175

60− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

176+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

177

62− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" change"},"finish_reason":null}]}

178+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

179

64− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" '"},"finish_reason":null}]}

180+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Summ"},"finish_reason":null}]}

181

66− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Hello"},"finish_reason":null}]}

182+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ar"},"finish_reason":null}]}

183

68− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

184+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ize"},"finish_reason":null}]}

185

70− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" World"},"finish_reason":null}]}

186+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

187

72− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"!'"},"finish_reason":null}]}

188+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" using"},"finish_reason":null}]}

189

74− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

190+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" mult"},"finish_reason":null}]}

191

76− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" '"},"finish_reason":null}]}

192+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ied"},"finish_reason":null}]}

193

78− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Hello"},"finish_reason":null}]}

194+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"it"},"finish_reason":null}]}

195

80− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

196+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

197

82− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Crush"},"finish_reason":null}]}

198+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" change"},"finish_reason":null}]}

199

84− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"!'"},"finish_reason":null}]}

200+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Hello"},"finish_reason":null}]}

201

86− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

202+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" World"},"finish_reason":null}]}

203

88− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" add"},"finish_reason":null}]}

204+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

205

90− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

206+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Hello"},"finish_reason":null}]}

207

92− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" comment"},"finish_reason":null}]}

208+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Crush"},"finish_reason":null}]}

209

94− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" '//"},"finish_reason":null}]}

210+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

211

96− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Greeting"},"finish_reason":null}]}

212+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" add"},"finish_reason":null}]}

213

98− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"'"},"finish_reason":null}]}

214+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" comment"},"finish_reason":null}]}

215

100− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" above"},"finish_reason":null}]}

216+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

101

217

102− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

218+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Title"},"finish_reason":null}]}

103

219

104− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" fmt"},"finish_reason":null}]}

220+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" could"},"finish_reason":null}]}

105

221

106− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".Println"},"finish_reason":null}]}

222+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

107

223

108− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

224+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

109

225

110− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

226+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

111

227

112− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

228+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Mult"},"finish_reason":null}]}

113

229

114− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".go"},"finish_reason":null}]}

230+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ied"},"finish_reason":null}]}

115

231

116− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

232+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"it"},"finish_reason":null}]}

117

233

118− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Title"},"finish_reason":null}]}

234+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

119

235

120− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" should"},"finish_reason":null}]}

236+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" change"},"finish_reason":null}]}

121

237

122− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

238+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" greeting"},"finish_reason":null}]}

123

239

124− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" short"},"finish_reason":null}]}

240+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

125

241

126− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" (\u003c"},"finish_reason":null}]}

242+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" add"},"finish_reason":null}]}

127

243

128− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

244+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" comment"},"finish_reason":null}]}

129

245

130− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

246+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

131

247

132− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"),"},"finish_reason":null}]}

248+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

133

249

134− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" summarize"},"finish_reason":null}]}

250+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

135

251

136− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

252+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

137

253

138− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Possibly"},"finish_reason":null}]}

254+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

139

255

140− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

256+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Mult"},"finish_reason":null}]}

141

257

142− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Edit"},"finish_reason":null}]}

258+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ied"},"finish_reason":null}]}

143

259

144− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

260+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"it"},"finish_reason":null}]}

145

261

146− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".go"},"finish_reason":null}]}

262+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

147

263

148− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

264+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" change"},"finish_reason":null}]}

149

265

150− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" change"},"finish_reason":null}]}

266+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" greeting"},"finish_reason":null}]}

151

267

152− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" greeting"},"finish_reason":null}]}

268+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

153

269

154− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

270+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" add"},"finish_reason":null}]}

155

271

156− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" add"},"finish_reason":null}]}

272+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" comment"},"finish_reason":null}]}

157

273

158− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" comment"},"finish_reason":null}]}

274+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

159

275

160− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

276+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" length"},"finish_reason":null}]}

161

277

162− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

278+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"?"},"finish_reason":null}]}

163

279

164− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

280+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Let's"},"finish_reason":null}]}

165

281

166− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

282+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

167

283

168− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

284+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

169

285

170− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Edit"},"finish_reason":null}]}

286+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Mult"},"finish_reason":null}]}

171

287

172− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

288+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ied"},"finish_reason":null}]}

173

289

174− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".go"},"finish_reason":null}]}

290+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"it"},"finish_reason":null}]}

175

291

176− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

292+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

177

293

178− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" change"},"finish_reason":null}]}

294+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"9"},"finish_reason":null}]}

179

295

180− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" greeting"},"finish_reason":null}]}

296+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

181

297

182− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

298+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" +"},"finish_reason":null}]}

183

299

184− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" add"},"finish_reason":null}]}

300+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" colon"},"finish_reason":null}]}

185

301

186− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" comment"},"finish_reason":null}]}

302+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

187

303

188− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

304+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

189

305

190− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ="},"finish_reason":null}]}

306+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

191

307

192− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" let's"},"finish_reason":null}]}

308+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" +"},"finish_reason":null}]}

193

309

194− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

310+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

195

311

196− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

312+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

197

313

198− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" E"},"finish_reason":null}]}

314+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

199

315

200− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

316+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

201

317

202− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

318+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ="},"finish_reason":null}]}

203

319

204− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")d"},"finish_reason":null}]}

320+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"11"},"finish_reason":null}]}

205

321

206− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"2"},"finish_reason":null}]}

322+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

207

323

208− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"i"},"finish_reason":null}]}

324+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

209

325

210− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"3"},"finish_reason":null}]}

326+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"change"},"finish_reason":null}]}

211

327

212− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"t"},"finish_reason":null}]}

328+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

213

329

214− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

330+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

215

331

216− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

332+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"6"},"finish_reason":null}]}

217

333

218− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"5"},"finish_reason":null}]}

334+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

219

335

220− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" m"},"finish_reason":null}]}

336+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" =\u003e"},"finish_reason":null}]}

221

337

222− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"6"},"finish_reason":null}]}

338+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

223

339

224− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"a"},"finish_reason":null}]}

340+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"17"},"finish_reason":null}]}

225

341

226− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"7"},"finish_reason":null}]}

342+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

227

343

228− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"i"},"finish_reason":null}]}

344+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

229

345

230− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"8"},"finish_reason":null}]}

346+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

231

347

232− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"n"},"finish_reason":null}]}

348+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

233

349

234− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"9"},"finish_reason":null}]}

350+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" =\u003e"},"finish_reason":null}]}

235

351

236− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

352+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"18"},"finish_reason":null}]}

237

353

238− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"10"},"finish_reason":null}]}

354+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

239

355

240− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"g"},"finish_reason":null}]}

356+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

241

357

242− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"11"},"finish_reason":null}]}

358+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"g"},"finish_reason":null}]}

243

359

244− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"o"},"finish_reason":null}]}

360+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"reeting"},"finish_reason":null}]}

245

361

246− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"12"},"finish_reason":null}]}

362+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

247

363

248− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".:"},"finish_reason":null}]}

364+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

249

365

250− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"13"},"finish_reason":null}]}

366+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"8"},"finish_reason":null}]}

251

367

252− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

368+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

253

369

254− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"14"},"finish_reason":null}]}

370+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" =\u003e"},"finish_reason":null}]}

255

371

256− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"c"},"finish_reason":null}]}

372+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"26"},"finish_reason":null}]}

257

373

258− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"15"},"finish_reason":null}]}

374+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

259

375

260− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"h"},"finish_reason":null}]}

376+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

261

377

262− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"16"},"finish_reason":null}]}

378+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

263

379

264− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"a"},"finish_reason":null}]}

380+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

265

381

266− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"17"},"finish_reason":null}]}

382+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" =\u003e"},"finish_reason":null}]}

267

383

268− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"n"},"finish_reason":null}]}

384+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"27"},"finish_reason":null}]}

269

385

270− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"18"},"finish_reason":null}]}

386+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

271

387

272− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"g"},"finish_reason":null}]}

388+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

273

389

274− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"19"},"finish_reason":null}]}

390+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"and"},"finish_reason":null}]}

275

391

276− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"e"},"finish_reason":null}]}

392+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

277

393

278− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"20"},"finish_reason":null}]}

394+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

279

395

280− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

396+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"3"},"finish_reason":null}]}

281

397

282− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"21"},"finish_reason":null}]}

398+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")=\u003e"},"finish_reason":null}]}

283

399

284− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"g"},"finish_reason":null}]}

400+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"30"},"finish_reason":null}]}

285

401

286− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"22"},"finish_reason":null}]}

402+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

287

403

288− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"r"},"finish_reason":null}]}

404+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

289

405

290− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"23"},"finish_reason":null}]}

406+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

291

407

292− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"e"},"finish_reason":null}]}

408+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"=\u003e"},"finish_reason":null}]}

293

409

294− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"24"},"finish_reason":null}]}

410+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"31"},"finish_reason":null}]}

295

411

296− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"e"},"finish_reason":null}]}

412+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

297

413

298− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"25"},"finish_reason":null}]}

414+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

299

415

300− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"t"},"finish_reason":null}]}

416+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"add"},"finish_reason":null}]}

301

417

302− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"26"},"finish_reason":null}]}

418+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"("},"finish_reason":null}]}

303

419

304− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"i"},"finish_reason":null}]}

420+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"3"},"finish_reason":null}]}

305

421

306− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"27"},"finish_reason":null}]}

422+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")=\u003e"},"finish_reason":null}]}

307

423

308− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"n"},"finish_reason":null}]}

424+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"34"},"finish_reason":null}]}

309

425

310− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"28"},"finish_reason":null}]}

426+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

311

427

312− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"g"},"finish_reason":null}]}

428+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

313

429

314− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"29"},"finish_reason":null}]}

430+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

315

431

316− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

432+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"=\u003e"},"finish_reason":null}]}

317

433

318− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"30"},"finish_reason":null}]}

434+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"35"},"finish_reason":null}]}

319

435

320− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"a"},"finish_reason":null}]}

436+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

321

437

322− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"31"},"finish_reason":null}]}

438+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

323

439

324− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"n"},"finish_reason":null}]}

440+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"comment"},"finish_reason":null}]}

325

441

326− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"32"},"finish_reason":null}]}

442+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"("},"finish_reason":null}]}

327

443

328− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"d"},"finish_reason":null}]}

444+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"7"},"finish_reason":null}]}

329

445

330− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"33"},"finish_reason":null}]}

446+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")=\u003e"},"finish_reason":null}]}

331

447

332− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

448+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"42"},"finish_reason":null}]}

333

449

334− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"34"},"finish_reason":null}]}

450+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

335

451

336− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"a"},"finish_reason":null}]}

452+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

337

453

338− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"35"},"finish_reason":null}]}

454+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

339

455

340− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"d"},"finish_reason":null}]}

456+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"42"},"finish_reason":null}]}

341

457

342− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"36"},"finish_reason":null}]}

458+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

343

459

344− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"d"},"finish_reason":null}]}

460+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

345

461

346− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"37"},"finish_reason":null}]}

462+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" okay"},"finish_reason":null}]}

347

463

348− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

464+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

349

465

350− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"38"},"finish_reason":null}]}

466+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" No"},"finish_reason":null}]}

351

467

352− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"c"},"finish_reason":null}]}

468+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" colon"},"finish_reason":null}]}

353

469

354− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"39"},"finish_reason":null}]}

470+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"?"},"finish_reason":null}]}

355

471

356− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"o"},"finish_reason":null}]}

472+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" It"},"finish_reason":null}]}

357

473

358− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"40"},"finish_reason":null}]}

474+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" says"},"finish_reason":null}]}

359

475

360− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"m"},"finish_reason":null}]}

476+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

361

477

362− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"41"},"finish_reason":null}]}

478+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" col"},"finish_reason":null}]}

363

479

364− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"m"},"finish_reason":null}]}

480+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ons"},"finish_reason":null}]}

365

481

366− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"42"},"finish_reason":null}]}

482+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

367

483

368− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"e"},"finish_reason":null}]}

484+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Actually"},"finish_reason":null}]}

369

485

370− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"43"},"finish_reason":null}]}

486+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" rule"},"finish_reason":null}]}

371

487

372− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"n"},"finish_reason":null}]}

488+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

373

489

374− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"44"},"finish_reason":null}]}

490+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Do"},"finish_reason":null}]}

375

491

376− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"t"},"finish_reason":null}]}

492+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" not"},"finish_reason":null}]}

377

493

378− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"45"},"finish_reason":null}]}

494+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" use"},"finish_reason":null}]}

379

495

380− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

496+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

381

497

382− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

498+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" or"},"finish_reason":null}]}

383

499

384− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

500+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" col"},"finish_reason":null}]}

385

501

386− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"45"},"finish_reason":null}]}

502+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ons"},"finish_reason":null}]}

387

503

388− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

504+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

389

505

390− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

506+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" So"},"finish_reason":null}]}

391

507

392− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" okay"},"finish_reason":null}]}

508+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" cannot"},"finish_reason":null}]}

393

509

394− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

510+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" use"},"finish_reason":null}]}

395

511

396− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Should"},"finish_reason":null}]}

512+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" colon"},"finish_reason":null}]}

397

513

398− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

514+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

399

515

400− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" one"},"finish_reason":null}]}

516+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Remove"},"finish_reason":null}]}

401

517

402− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

518+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" colon"},"finish_reason":null}]}

403

519

404− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

520+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

405

521

406− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

522+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Title"},"finish_reason":null}]}

407

523

408− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

524+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

409

525

410− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

526+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

411

527

412− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Good"},"finish_reason":null}]}

528+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Mult"},"finish_reason":null}]}

413

529

414− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

530+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ied"},"finish_reason":null}]}

415

531

416− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Edit"},"finish_reason":null}]}

532+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"it"},"finish_reason":null}]}

417

533

418− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" main"},"finish_reason":null}]}

534+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" change"},"finish_reason":null}]}

419

535

420− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":".go"},"finish_reason":null}]}

536+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" greeting"},"finish_reason":null}]}

421

537

422− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":":"},"finish_reason":null}]}

538+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

423

539

424− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" change"},"finish_reason":null}]}

540+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" add"},"finish_reason":null}]}

425

541

426− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" greeting"},"finish_reason":null}]}

542+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" comment"},"finish_reason":null}]}

427

543

428− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" and"},"finish_reason":null}]}

544+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

429

545

430− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" add"},"finish_reason":null}]}

546+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

431

547

432− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" comment"},"finish_reason":null}]}

548+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

433

549

434− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

550+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Mult"},"finish_reason":null}]}

435

551

436− data: {"id":"chatcmpl-c22d983eebe935bed8861df2","object":"chat.completion.chunk","created":1779218903,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":235,"completion_tokens":212,"total_tokens":447,"completion_tokens_details":{"reasoning_tokens":148},"cost":{"usd":0.0001675,"hypercredits":0.00335},"remaining":{"hypercredits":0}}}

552+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ied"},"finish_reason":null}]}

437

553

438− data: [DONE]

554+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"it"},"finish_reason":null}]}

439

555

440− headers:

556+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

441− Content-Type:

557+

442− - text/event-stream

558+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"9"},"finish_reason":null}]}

443− status: 200 OK

559+

444− code: 200

560+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

445− duration: 652.738167ms

561+

446−- id: 1

562+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" +"},"finish_reason":null}]}

447− request:

563+

448− proto: HTTP/1.1

564+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

449− proto_major: 1

565+

450− proto_minor: 1

566+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

451− content_length: 34043

567+

452− host: ""

568+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

453− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use multiedit to change 'Hello, World!' to 'Hello, Crush!' and add a comment '// Greeting' above the fmt.Println line in main.go\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

569+

454− headers:

570+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ="},"finish_reason":null}]}

455− Accept:

456− - application/json

457− Content-Type:

458− - application/json

459− User-Agent:

460− - Charm-Crush/devel (https://charm.land/crush)

461− url: https://hyper.charm.land/v1/chat/completions

462− method: POST

463− response:

464− proto: HTTP/2.0

465− proto_major: 2

466− proto_minor: 0

467− content_length: -1

468− uncompressed: true

469− body: |+

470− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

471

571

472− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"Let"},"finish_reason":null}]}

572+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"10"},"finish_reason":null}]}

473

573

474− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" me"},"finish_reason":null}]}

574+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

475

575

476− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" first read"},"finish_reason":null}]}

576+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" change"},"finish_reason":null}]}

477

577

478− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" the main"},"finish_reason":null}]}

578+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

479

579

480− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":".go file"},"finish_reason":null}]}

580+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"6"},"finish_reason":null}]}

481

581

482− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" to understand"},"finish_reason":null}]}

582+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

483

583

484− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" its"},"finish_reason":null}]}

584+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" =\u003e"},"finish_reason":null}]}

485

585

486− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" current content"},"finish_reason":null}]}

586+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"16"},"finish_reason":null}]}

487

587

488− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

588+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

489

589

490− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_67520d79d7f24633bd817fa6","type":"function","index":0,"function":{"name":"view","arguments":""}}]},"finish_reason":null}]}

590+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

491

591

492− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

592+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

493

593

494− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

594+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

495

595

496− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"file_path\": "}}]},"finish_reason":null}]}

596+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" =\u003e"},"finish_reason":null}]}

497

597

498− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"/tmp"}}]},"finish_reason":null}]}

598+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"17"},"finish_reason":null}]}

499

599

500− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/c"}}]},"finish_reason":null}]}

600+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

501

601

502− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"rush-test"}}]},"finish_reason":null}]}

602+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" greeting"},"finish_reason":null}]}

503

603

504− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/Test"}}]},"finish_reason":null}]}

604+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

505

605

506− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Coder"}}]},"finish_reason":null}]}

606+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"8"},"finish_reason":null}]}

507

607

508− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Agent"}}]},"finish_reason":null}]}

608+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")=\u003e"},"finish_reason":null}]}

509

609

510− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/glm"}}]},"finish_reason":null}]}

610+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"25"},"finish_reason":null}]}

511

611

512− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-5."}}]},"finish_reason":null}]}

612+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

513

613

514− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"1/mult"}}]},"finish_reason":null}]}

614+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

515

615

516− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"ied"}}]},"finish_reason":null}]}

616+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

517

617

518− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"it_tool"}}]},"finish_reason":null}]}

618+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

519

619

520− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/main.go\""}}]},"finish_reason":null}]}

620+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" =\u003e"},"finish_reason":null}]}

521

621

522− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"}"}}]},"finish_reason":null}]}

622+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"26"},"finish_reason":null}]}

523

623

524− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

624+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

525

625

626+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

627+

628+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

629+

630+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"3"},"finish_reason":null}]}

631+

632+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")=\u003e"},"finish_reason":null}]}

633+

634+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"29"},"finish_reason":null}]}

635+

636+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

637+

638+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

639+

640+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

641+

642+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

643+

644+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"=\u003e"},"finish_reason":null}]}

645+

646+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"30"},"finish_reason":null}]}

647+

648+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

649+

650+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" add"},"finish_reason":null}]}

651+

652+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

653+

654+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"3"},"finish_reason":null}]}

655+

656+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")=\u003e"},"finish_reason":null}]}

657+

658+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"33"},"finish_reason":null}]}

659+

660+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

661+

662+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

663+

664+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

665+

666+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

667+

668+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"=\u003e"},"finish_reason":null}]}

669+

670+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"34"},"finish_reason":null}]}

671+

672+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

673+

674+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" comment"},"finish_reason":null}]}

675+

676+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

677+

678+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"7"},"finish_reason":null}]}

679+

680+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")=\u003e"},"finish_reason":null}]}

681+

682+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"41"},"finish_reason":null}]}

683+

684+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

685+

686+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

687+

688+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"41"},"finish_reason":null}]}

689+

690+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

691+

692+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

693+

694+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" okay"},"finish_reason":null}]}

695+

696+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

697+

698+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" No"},"finish_reason":null}]}

699+

700+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" colon"},"finish_reason":null}]}

701+

702+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

703+

704+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Return"},"finish_reason":null}]}

705+

706+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" that"},"finish_reason":null}]}

707+

708+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

709+

710+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

711+

712+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Mult"},"finish_reason":null}]}

713+

714+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"ied"},"finish_reason":null}]}

715+

716+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"it"},"finish_reason":null}]}

717+

718+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" change"},"finish_reason":null}]}

719+

720+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" greeting"},"finish_reason":null}]}

721+

722+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" and"},"finish_reason":null}]}

723+

724+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" add"},"finish_reason":null}]}

725+

726+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" comment"},"finish_reason":null}]}

727+

728+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

729+

526− data: {"id":"chatcmpl-714ee23cbaa8f78c4e111f6c","object":"chat.completion.chunk","created":1779218906,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":736,"completion_tokens":45,"total_tokens":7885,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":14},"cost":{"usd":0.0013636,"hypercredits":0.027271999999999998},"remaining":{"hypercredits":0}}}

730+ data: {"id":"chatcmpl-f0ea2851314350c1bab8d6ee","object":"chat.completion.chunk","created":1780069697,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":236,"completion_tokens":301,"total_tokens":537,"completion_tokens_details":{"reasoning_tokens":284},"cost":{"usd":0.00039933,"hypercredits":0.0079866},"remaining":{"hypercredits":0}}}

527

731

528 data: [DONE]

732 data: [DONE]

529

733

@@ -532,15 +736,15 @@ interactions:

532 - text/event-stream

736 - text/event-stream

533 status: 200 OK

737 status: 200 OK

534 code: 200

738 code: 200

535− duration: 3.083916208s

739+ duration: 394.4095ms

536- id: 2

740- id: 2

537 request:

741 request:

538 proto: HTTP/1.1

742 proto: HTTP/1.1

539 proto_major: 1

743 proto_major: 1

540 proto_minor: 1

744 proto_minor: 1

541− content_length: 34589

745+ content_length: 34570

542 host: ""

746 host: ""

543− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use multiedit to change 'Hello, World!' to 'Hello, Crush!' and add a comment '// Greeting' above the fmt.Println line in main.go\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_67520d79d7f24633bd817fa6\",\"function\":{\"arguments\":\"{\\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool/main.go\\\"}\",\"name\":\"view\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"Let me first read the main.go file to understand its current content.\"},{\"content\":\"<file>\\n 1|package main\\n 2|\\n 3|import \\\"fmt\\\"\\n 4|\\n 5|func main() {\\n 6|\\tfmt.Println(\\\"Hello, World!\\\")\\n 7|}\\n 8|\\n</file>\\n\",\"tool_call_id\":\"call_67520d79d7f24633bd817fa6\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

747+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use multiedit to change 'Hello, World!' to 'Hello, Crush!' and add a comment '// Greeting' above the fmt.Println line in main.go\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_4BvAOulZ4vPbT3o9XRDa7mrF\",\"function\":{\"arguments\":\"{\\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool/main.go\\\"}\",\"name\":\"view\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"Let me first read the main.go file to understand its current content.\"},{\"content\":\"<file>\\n 1|package main\\n 2|\\n 3|import \\\"fmt\\\"\\n 4|\\n 5|func main() {\\n 6|\\tfmt.Println(\\\"Hello, World!\\\")\\n 7|}\\n 8|\\n</file>\\n\",\"tool_call_id\":\"call_4BvAOulZ4vPbT3o9XRDa7mrF\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

544 headers:

748 headers:

545 Accept:

749 Accept:

546 - application/json

750 - application/json

@@ -557,101 +761,175 @@ interactions:

557 content_length: -1

761 content_length: -1

558 uncompressed: true

762 uncompressed: true

559 body: |+

763 body: |+

560− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

764+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

765+

766+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_aHnPYHta9B2b9HfMoNV5jxjf","type":"function","index":0,"function":{"name":"multiedit","arguments":""}}]},"finish_reason":null}]}

561

767

562− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_375fbb883f564b398a9670c3","type":"function","index":0,"function":{"name":"multiedit","arguments":""}}]},"finish_reason":null}]}

768+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

563

769

564− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

770+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

565

771

566− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

772+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"edits\": "}}]},"finish_reason":null}]}

567

773

568− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"edits\": "}}]},"finish_reason":null}]}

774+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"["}}]},"finish_reason":null}]}

569

775

570− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"[{\""}}]},"finish_reason":null}]}

776+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{\""}}]},"finish_reason":null}]}

571

777

572− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"new_string"}}]},"finish_reason":null}]}

778+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"new"}}]},"finish_reason":null}]}

573

779

574− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\": \""}}]},"finish_reason":null}]}

780+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_string"}}]},"finish_reason":null}]}

575

781

576− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"fmt.Println"}}]},"finish_reason":null}]}

782+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\":"}}]},"finish_reason":null}]}

577

783

578− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"(\\\"Hello, Crush"}}]},"finish_reason":null}]}

784+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" \""}}]},"finish_reason":null}]}

579

785

580− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"!\\\")"}}]},"finish_reason":null}]}

786+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Hello"}}]},"finish_reason":null}]}

581

787

582− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\", \""}}]},"finish_reason":null}]}

788+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":","}}]},"finish_reason":null}]}

583

789

584− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"old_string"}}]},"finish_reason":null}]}

790+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" Crush"}}]},"finish_reason":null}]}

585

791

586− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\": \""}}]},"finish_reason":null}]}

792+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"!\","}}]},"finish_reason":null}]}

587

793

588− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"fmt.Println(\\\""}}]},"finish_reason":null}]}

794+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" \""}}]},"finish_reason":null}]}

589

795

590− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Hello, World"}}]},"finish_reason":null}]}

796+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"old"}}]},"finish_reason":null}]}

591

797

592− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"!\\\")"}}]},"finish_reason":null}]}

798+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_string"}}]},"finish_reason":null}]}

593

799

594− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"}, {\""}}]},"finish_reason":null}]}

800+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\":"}}]},"finish_reason":null}]}

595

801

596− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"new_string\":"}}]},"finish_reason":null}]}

802+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" \""}}]},"finish_reason":null}]}

597

803

598− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" \"// G"}}]},"finish_reason":null}]}

804+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Hello"}}]},"finish_reason":null}]}

599

805

600− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"reeting\\n\\t"}}]},"finish_reason":null}]}

806+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":","}}]},"finish_reason":null}]}

601

807

602− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"fmt.Println(\\\""}}]},"finish_reason":null}]}

808+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" World"}}]},"finish_reason":null}]}

603

809

604− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Hello, Crush"}}]},"finish_reason":null}]}

810+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"!\""}}]},"finish_reason":null}]}

605

811

606− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"!\\\")"}}]},"finish_reason":null}]}

812+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"},"}}]},"finish_reason":null}]}

607

813

608− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\", \"old"}}]},"finish_reason":null}]}

814+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" {\""}}]},"finish_reason":null}]}

609

815

610− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_string\": \""}}]},"finish_reason":null}]}

816+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"new"}}]},"finish_reason":null}]}

611

817

612− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"fmt.Println(\\\""}}]},"finish_reason":null}]}

818+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_string"}}]},"finish_reason":null}]}

613

819

614− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Hello, Crush"}}]},"finish_reason":null}]}

820+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\":"}}]},"finish_reason":null}]}

615

821

616− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"!\\\")"}}]},"finish_reason":null}]}

822+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" \"\\"}}]},"finish_reason":null}]}

617

823

618− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"}]"}}]},"finish_reason":null}]}

824+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"t"}}]},"finish_reason":null}]}

619

825

620− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":", "}}]},"finish_reason":null}]}

826+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"//"}}]},"finish_reason":null}]}

621

827

622− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"file_path\": "}}]},"finish_reason":null}]}

828+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" G"}}]},"finish_reason":null}]}

623

829

624− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"/tmp/c"}}]},"finish_reason":null}]}

830+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"reeting"}}]},"finish_reason":null}]}

625

831

626− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"rush"}}]},"finish_reason":null}]}

832+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\\n"}}]},"finish_reason":null}]}

627

833

628− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-test"}}]},"finish_reason":null}]}

834+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\\t"}}]},"finish_reason":null}]}

629

835

630− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/Test"}}]},"finish_reason":null}]}

836+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"fmt"}}]},"finish_reason":null}]}

631

837

632− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Coder"}}]},"finish_reason":null}]}

838+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".Println"}}]},"finish_reason":null}]}

633

839

634− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Agent"}}]},"finish_reason":null}]}

840+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"(\\\""}}]},"finish_reason":null}]}

635

841

636− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/glm"}}]},"finish_reason":null}]}

842+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Hello"}}]},"finish_reason":null}]}

637

843

638− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-5."}}]},"finish_reason":null}]}

844+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":","}}]},"finish_reason":null}]}

639

845

640− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"1/m"}}]},"finish_reason":null}]}

846+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" Crush"}}]},"finish_reason":null}]}

641

847

642− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"ultied"}}]},"finish_reason":null}]}

848+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"!\\"}}]},"finish_reason":null}]}

643

849

644− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"it"}}]},"finish_reason":null}]}

850+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\")"}}]},"finish_reason":null}]}

645

851

646− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_tool/main"}}]},"finish_reason":null}]}

852+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\","}}]},"finish_reason":null}]}

647

853

648− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".go\""}}]},"finish_reason":null}]}

854+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" \""}}]},"finish_reason":null}]}

649

855

650− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"}"}}]},"finish_reason":null}]}

856+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"old"}}]},"finish_reason":null}]}

651

857

652− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

858+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_string"}}]},"finish_reason":null}]}

653

859

860+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\":"}}]},"finish_reason":null}]}

861+

862+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" \"\\"}}]},"finish_reason":null}]}

863+

864+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"tf"}}]},"finish_reason":null}]}

865+

866+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"mt"}}]},"finish_reason":null}]}

867+

868+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".Println"}}]},"finish_reason":null}]}

869+

870+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"(\\\""}}]},"finish_reason":null}]}

871+

872+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Hello"}}]},"finish_reason":null}]}

873+

874+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":","}}]},"finish_reason":null}]}

875+

876+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" Crush"}}]},"finish_reason":null}]}

877+

878+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"!\\"}}]},"finish_reason":null}]}

879+

880+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\")"}}]},"finish_reason":null}]}

881+

882+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"}"}}]},"finish_reason":null}]}

883+

884+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"]"}}]},"finish_reason":null}]}

885+

886+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":", \"file_path\": "}}]},"finish_reason":null}]}

887+

888+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"/tmp"}}]},"finish_reason":null}]}

889+

890+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/c"}}]},"finish_reason":null}]}

891+

892+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"rush"}}]},"finish_reason":null}]}

893+

894+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-test"}}]},"finish_reason":null}]}

895+

896+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/Test"}}]},"finish_reason":null}]}

897+

898+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Coder"}}]},"finish_reason":null}]}

899+

900+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Agent"}}]},"finish_reason":null}]}

901+

902+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/g"}}]},"finish_reason":null}]}

903+

904+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"lm"}}]},"finish_reason":null}]}

905+

906+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-"}}]},"finish_reason":null}]}

907+

908+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"5"}}]},"finish_reason":null}]}

909+

910+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"."}}]},"finish_reason":null}]}

911+

912+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"1"}}]},"finish_reason":null}]}

913+

914+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/m"}}]},"finish_reason":null}]}

915+

916+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"ult"}}]},"finish_reason":null}]}

917+

918+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"ied"}}]},"finish_reason":null}]}

919+

920+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"it"}}]},"finish_reason":null}]}

921+

922+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_tool"}}]},"finish_reason":null}]}

923+

924+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/main"}}]},"finish_reason":null}]}

925+

926+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".go"}}]},"finish_reason":null}]}

927+

928+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"}"}}]},"finish_reason":null}]}

929+

930+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

931+

654− data: {"id":"chatcmpl-894599274b43bf493bdc74ea","object":"chat.completion.chunk","created":1779218910,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":123,"completion_tokens":101,"total_tokens":8032,"prompt_tokens_details":{"cached_tokens":7808},"completion_tokens_details":{"reasoning_tokens":100},"cost":{"usd":0.0010689000000000002,"hypercredits":0.021378000000000005},"remaining":{"hypercredits":0}}}

932+ data: {"id":"chatcmpl-4c94d5eb32ed6ae8c840e50c","object":"chat.completion.chunk","created":1780069698,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":118,"completion_tokens":95,"total_tokens":8021,"prompt_tokens_details":{"cached_tokens":7808},"cost":{"usd":0.00055358,"hypercredits":0.0110716},"remaining":{"hypercredits":0}}}

655

933

656 data: [DONE]

934 data: [DONE]

657

935

@@ -660,15 +938,15 @@ interactions:

660 - text/event-stream

938 - text/event-stream

661 status: 200 OK

939 status: 200 OK

662 code: 200

940 code: 200

663− duration: 2.97328875s

941+ duration: 521.977833ms

664- id: 3

942- id: 3

665 request:

943 request:

666 proto: HTTP/1.1

944 proto: HTTP/1.1

667 proto_major: 1

945 proto_major: 1

668 proto_minor: 1

946 proto_minor: 1

669− content_length: 35277

947+ content_length: 35222

670 host: ""

948 host: ""

671− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use multiedit to change 'Hello, World!' to 'Hello, Crush!' and add a comment '// Greeting' above the fmt.Println line in main.go\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_67520d79d7f24633bd817fa6\",\"function\":{\"arguments\":\"{\\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool/main.go\\\"}\",\"name\":\"view\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"Let me first read the main.go file to understand its current content.\"},{\"content\":\"<file>\\n 1|package main\\n 2|\\n 3|import \\\"fmt\\\"\\n 4|\\n 5|func main() {\\n 6|\\tfmt.Println(\\\"Hello, World!\\\")\\n 7|}\\n 8|\\n</file>\\n\",\"tool_call_id\":\"call_67520d79d7f24633bd817fa6\",\"role\":\"tool\"},{\"tool_calls\":[{\"id\":\"call_375fbb883f564b398a9670c3\",\"function\":{\"arguments\":\"{\\\"edits\\\": [{\\\"new_string\\\": \\\"fmt.Println(\\\\\\\"Hello, Crush!\\\\\\\")\\\", \\\"old_string\\\": \\\"fmt.Println(\\\\\\\"Hello, World!\\\\\\\")\\\"}, {\\\"new_string\\\": \\\"// Greeting\\\\n\\\\tfmt.Println(\\\\\\\"Hello, Crush!\\\\\\\")\\\", \\\"old_string\\\": \\\"fmt.Println(\\\\\\\"Hello, Crush!\\\\\\\")\\\"}], \\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool/main.go\\\"}\",\"name\":\"multiedit\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"\"},{\"content\":\"<result>\\nApplied 2 edits to file: /tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool/main.go\\n</result>\\n\",\"tool_call_id\":\"call_375fbb883f564b398a9670c3\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

949+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use multiedit to change 'Hello, World!' to 'Hello, Crush!' and add a comment '// Greeting' above the fmt.Println line in main.go\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_4BvAOulZ4vPbT3o9XRDa7mrF\",\"function\":{\"arguments\":\"{\\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool/main.go\\\"}\",\"name\":\"view\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"Let me first read the main.go file to understand its current content.\"},{\"content\":\"<file>\\n 1|package main\\n 2|\\n 3|import \\\"fmt\\\"\\n 4|\\n 5|func main() {\\n 6|\\tfmt.Println(\\\"Hello, World!\\\")\\n 7|}\\n 8|\\n</file>\\n\",\"tool_call_id\":\"call_4BvAOulZ4vPbT3o9XRDa7mrF\",\"role\":\"tool\"},{\"tool_calls\":[{\"id\":\"call_aHnPYHta9B2b9HfMoNV5jxjf\",\"function\":{\"arguments\":\"{\\\"edits\\\": [{\\\"new_string\\\": \\\"Hello, Crush!\\\", \\\"old_string\\\": \\\"Hello, World!\\\"}, {\\\"new_string\\\": \\\"\\\\t// Greeting\\\\n\\\\tfmt.Println(\\\\\\\"Hello, Crush!\\\\\\\")\\\", \\\"old_string\\\": \\\"\\\\tfmt.Println(\\\\\\\"Hello, Crush!\\\\\\\")\\\"}], \\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool/main.go\\\"}\",\"name\":\"multiedit\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"\"},{\"content\":\"<result>\\nApplied 2 edits to file: /tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool/main.go\\n</result>\\n\",\"tool_call_id\":\"call_aHnPYHta9B2b9HfMoNV5jxjf\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

672 headers:

950 headers:

673 Accept:

951 Accept:

674 - application/json

952 - application/json

@@ -685,43 +963,61 @@ interactions:

685 content_length: -1

963 content_length: -1

686 uncompressed: true

964 uncompressed: true

687 body: |+

965 body: |+

688− data: {"id":"chatcmpl-6816fb2f0f434f5db4f29bf0","object":"chat.completion.chunk","created":1779218913,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

966+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

967+

968+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_Qp5BlRVFac2IqxDa86BHkiPX","type":"function","index":0,"function":{"name":"view","arguments":""}}]},"finish_reason":null}]}

969+

970+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

971+

972+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

973+

974+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"file_path\": "}}]},"finish_reason":null}]}

689

975

690− data: {"id":"chatcmpl-6816fb2f0f434f5db4f29bf0","object":"chat.completion.chunk","created":1779218913,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_e720ab9f40a941e8adb348cc","type":"function","index":0,"function":{"name":"view","arguments":""}}]},"finish_reason":null}]}

976+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"/tmp"}}]},"finish_reason":null}]}

691

977

692− data: {"id":"chatcmpl-6816fb2f0f434f5db4f29bf0","object":"chat.completion.chunk","created":1779218913,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

978+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/c"}}]},"finish_reason":null}]}

693

979

694− data: {"id":"chatcmpl-6816fb2f0f434f5db4f29bf0","object":"chat.completion.chunk","created":1779218913,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

980+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"rush"}}]},"finish_reason":null}]}

695

981

696− data: {"id":"chatcmpl-6816fb2f0f434f5db4f29bf0","object":"chat.completion.chunk","created":1779218913,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"file_path\": "}}]},"finish_reason":null}]}

982+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-test"}}]},"finish_reason":null}]}

697

983

698− data: {"id":"chatcmpl-6816fb2f0f434f5db4f29bf0","object":"chat.completion.chunk","created":1779218913,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"/tmp/c"}}]},"finish_reason":null}]}

984+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/Test"}}]},"finish_reason":null}]}

699

985

700− data: {"id":"chatcmpl-6816fb2f0f434f5db4f29bf0","object":"chat.completion.chunk","created":1779218913,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"rush-test"}}]},"finish_reason":null}]}

986+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Coder"}}]},"finish_reason":null}]}

701

987

702− data: {"id":"chatcmpl-6816fb2f0f434f5db4f29bf0","object":"chat.completion.chunk","created":1779218913,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/Test"}}]},"finish_reason":null}]}

988+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Agent"}}]},"finish_reason":null}]}

703

989

704− data: {"id":"chatcmpl-6816fb2f0f434f5db4f29bf0","object":"chat.completion.chunk","created":1779218913,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"CoderAgent"}}]},"finish_reason":null}]}

990+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/g"}}]},"finish_reason":null}]}

705

991

706− data: {"id":"chatcmpl-6816fb2f0f434f5db4f29bf0","object":"chat.completion.chunk","created":1779218913,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/glm-"}}]},"finish_reason":null}]}

992+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"lm"}}]},"finish_reason":null}]}

707

993

708− data: {"id":"chatcmpl-6816fb2f0f434f5db4f29bf0","object":"chat.completion.chunk","created":1779218913,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"5."}}]},"finish_reason":null}]}

994+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-"}}]},"finish_reason":null}]}

709

995

710− data: {"id":"chatcmpl-6816fb2f0f434f5db4f29bf0","object":"chat.completion.chunk","created":1779218913,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"1/mult"}}]},"finish_reason":null}]}

996+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"5"}}]},"finish_reason":null}]}

711

997

712− data: {"id":"chatcmpl-6816fb2f0f434f5db4f29bf0","object":"chat.completion.chunk","created":1779218913,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"ied"}}]},"finish_reason":null}]}

998+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"."}}]},"finish_reason":null}]}

713

999

714− data: {"id":"chatcmpl-6816fb2f0f434f5db4f29bf0","object":"chat.completion.chunk","created":1779218913,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"it_tool"}}]},"finish_reason":null}]}

1000+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"1"}}]},"finish_reason":null}]}

715

1001

716− data: {"id":"chatcmpl-6816fb2f0f434f5db4f29bf0","object":"chat.completion.chunk","created":1779218913,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/main.go"}}]},"finish_reason":null}]}

1002+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/m"}}]},"finish_reason":null}]}

717

1003

718− data: {"id":"chatcmpl-6816fb2f0f434f5db4f29bf0","object":"chat.completion.chunk","created":1779218913,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\""}}]},"finish_reason":null}]}

1004+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"ult"}}]},"finish_reason":null}]}

719

1005

720− data: {"id":"chatcmpl-6816fb2f0f434f5db4f29bf0","object":"chat.completion.chunk","created":1779218913,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"}"}}]},"finish_reason":null}]}

1006+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"ied"}}]},"finish_reason":null}]}

721

1007

722− data: {"id":"chatcmpl-6816fb2f0f434f5db4f29bf0","object":"chat.completion.chunk","created":1779218913,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

1008+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"it"}}]},"finish_reason":null}]}

723

1009

724− data: {"id":"chatcmpl-6816fb2f0f434f5db4f29bf0","object":"chat.completion.chunk","created":1779218913,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":199,"completion_tokens":31,"total_tokens":8102,"prompt_tokens_details":{"cached_tokens":7872},"completion_tokens_details":{"reasoning_tokens":30},"cost":{"usd":0.0005669,"hypercredits":0.011337999999999999},"remaining":{"hypercredits":0}}}

1010+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_tool"}}]},"finish_reason":null}]}

1011+

1012+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/main"}}]},"finish_reason":null}]}

1013+

1014+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".go"}}]},"finish_reason":null}]}

1015+

1016+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"}"}}]},"finish_reason":null}]}

1017+

1018+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

1019+

1020+ data: {"id":"chatcmpl-731f796a13fe54cfbf8196d7","object":"chat.completion.chunk","created":1780069699,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":60,"completion_tokens":31,"total_tokens":8091,"prompt_tokens_details":{"cached_tokens":8000},"cost":{"usd":0.00020880000000000003,"hypercredits":0.004176},"remaining":{"hypercredits":0}}}

725

1021

726 data: [DONE]

1022 data: [DONE]

727

1023

@@ -730,15 +1026,15 @@ interactions:

730 - text/event-stream

1026 - text/event-stream

731 status: 200 OK

1027 status: 200 OK

732 code: 200

1028 code: 200

733− duration: 1.9770875s

1029+ duration: 895.18625ms

734- id: 4

1030- id: 4

735 request:

1031 request:

736 proto: HTTP/1.1

1032 proto: HTTP/1.1

737 proto_major: 1

1033 proto_major: 1

738 proto_minor: 1

1034 proto_minor: 1

739− content_length: 35776

1035+ content_length: 35721

740 host: ""

1036 host: ""

741− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use multiedit to change 'Hello, World!' to 'Hello, Crush!' and add a comment '// Greeting' above the fmt.Println line in main.go\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_67520d79d7f24633bd817fa6\",\"function\":{\"arguments\":\"{\\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool/main.go\\\"}\",\"name\":\"view\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"Let me first read the main.go file to understand its current content.\"},{\"content\":\"<file>\\n 1|package main\\n 2|\\n 3|import \\\"fmt\\\"\\n 4|\\n 5|func main() {\\n 6|\\tfmt.Println(\\\"Hello, World!\\\")\\n 7|}\\n 8|\\n</file>\\n\",\"tool_call_id\":\"call_67520d79d7f24633bd817fa6\",\"role\":\"tool\"},{\"tool_calls\":[{\"id\":\"call_375fbb883f564b398a9670c3\",\"function\":{\"arguments\":\"{\\\"edits\\\": [{\\\"new_string\\\": \\\"fmt.Println(\\\\\\\"Hello, Crush!\\\\\\\")\\\", \\\"old_string\\\": \\\"fmt.Println(\\\\\\\"Hello, World!\\\\\\\")\\\"}, {\\\"new_string\\\": \\\"// Greeting\\\\n\\\\tfmt.Println(\\\\\\\"Hello, Crush!\\\\\\\")\\\", \\\"old_string\\\": \\\"fmt.Println(\\\\\\\"Hello, Crush!\\\\\\\")\\\"}], \\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool/main.go\\\"}\",\"name\":\"multiedit\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"\"},{\"content\":\"<result>\\nApplied 2 edits to file: /tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool/main.go\\n</result>\\n\",\"tool_call_id\":\"call_375fbb883f564b398a9670c3\",\"role\":\"tool\"},{\"tool_calls\":[{\"id\":\"call_e720ab9f40a941e8adb348cc\",\"function\":{\"arguments\":\"{\\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool/main.go\\\"}\",\"name\":\"view\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"\"},{\"content\":\"<file>\\n 1|package main\\n 2|\\n 3|import \\\"fmt\\\"\\n 4|\\n 5|func main() {\\n 6|\\t// Greeting\\n 7|\\tfmt.Println(\\\"Hello, Crush!\\\")\\n 8|}\\n 9|\\n</file>\\n\",\"tool_call_id\":\"call_e720ab9f40a941e8adb348cc\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

1037+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use multiedit to change 'Hello, World!' to 'Hello, Crush!' and add a comment '// Greeting' above the fmt.Println line in main.go\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_4BvAOulZ4vPbT3o9XRDa7mrF\",\"function\":{\"arguments\":\"{\\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool/main.go\\\"}\",\"name\":\"view\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"Let me first read the main.go file to understand its current content.\"},{\"content\":\"<file>\\n 1|package main\\n 2|\\n 3|import \\\"fmt\\\"\\n 4|\\n 5|func main() {\\n 6|\\tfmt.Println(\\\"Hello, World!\\\")\\n 7|}\\n 8|\\n</file>\\n\",\"tool_call_id\":\"call_4BvAOulZ4vPbT3o9XRDa7mrF\",\"role\":\"tool\"},{\"tool_calls\":[{\"id\":\"call_aHnPYHta9B2b9HfMoNV5jxjf\",\"function\":{\"arguments\":\"{\\\"edits\\\": [{\\\"new_string\\\": \\\"Hello, Crush!\\\", \\\"old_string\\\": \\\"Hello, World!\\\"}, {\\\"new_string\\\": \\\"\\\\t// Greeting\\\\n\\\\tfmt.Println(\\\\\\\"Hello, Crush!\\\\\\\")\\\", \\\"old_string\\\": \\\"\\\\tfmt.Println(\\\\\\\"Hello, Crush!\\\\\\\")\\\"}], \\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool/main.go\\\"}\",\"name\":\"multiedit\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"\"},{\"content\":\"<result>\\nApplied 2 edits to file: /tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool/main.go\\n</result>\\n\",\"tool_call_id\":\"call_aHnPYHta9B2b9HfMoNV5jxjf\",\"role\":\"tool\"},{\"tool_calls\":[{\"id\":\"call_Qp5BlRVFac2IqxDa86BHkiPX\",\"function\":{\"arguments\":\"{\\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/multiedit_tool/main.go\\\"}\",\"name\":\"view\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"\"},{\"content\":\"<file>\\n 1|package main\\n 2|\\n 3|import \\\"fmt\\\"\\n 4|\\n 5|func main() {\\n 6|\\t// Greeting\\n 7|\\tfmt.Println(\\\"Hello, Crush!\\\")\\n 8|}\\n 9|\\n</file>\\n\",\"tool_call_id\":\"call_Qp5BlRVFac2IqxDa86BHkiPX\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

742 headers:

1038 headers:

743 Accept:

1039 Accept:

744 - application/json

1040 - application/json

@@ -755,47 +1051,63 @@ interactions:

755 content_length: -1

1051 content_length: -1

756 uncompressed: true

1052 uncompressed: true

757 body: |+

1053 body: |+

758− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

1054+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

1055+

1056+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Done"},"finish_reason":null}]}

1057+

1058+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"."},"finish_reason":null}]}

1059+

1060+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" Updated"},"finish_reason":null}]}

1061+

1062+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" greeting"},"finish_reason":null}]}

1063+

1064+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" to"},"finish_reason":null}]}

1065+

1066+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" \""},"finish_reason":null}]}

1067+

1068+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Hello"},"finish_reason":null}]}

1069+

1070+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":","},"finish_reason":null}]}

759

1071

760− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Done"},"finish_reason":null}]}

1072+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" Crush"},"finish_reason":null}]}

761

1073

762− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":". Two"},"finish_reason":null}]}

1074+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"!\""},"finish_reason":null}]}

763

1075

764− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" edits"},"finish_reason":null}]}

1076+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" and"},"finish_reason":null}]}

765

1077

766− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" applied: string"},"finish_reason":null}]}

1078+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" added"},"finish_reason":null}]}

767

1079

768− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" changed to"},"finish_reason":null}]}

1080+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

769

1081

770− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `\""},"finish_reason":null}]}

1082+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"//"},"finish_reason":null}]}

771

1083

772− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Hello,"},"finish_reason":null}]}

1084+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" G"},"finish_reason":null}]}

773

1085

774− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" Crush!\"`"},"finish_reason":null}]}

1086+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"reeting"},"finish_reason":null}]}

775

1087

776− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" and `"},"finish_reason":null}]}

1088+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`"},"finish_reason":null}]}

777

1089

778− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"// G"},"finish_reason":null}]}

1090+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" comment"},"finish_reason":null}]}

779

1091

780− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"reeting"},"finish_reason":null}]}

1092+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" above"},"finish_reason":null}]}

781

1093

782− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"` comment"},"finish_reason":null}]}

1094+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" the"},"finish_reason":null}]}

783

1095

784− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" added above"},"finish_reason":null}]}

1096+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

785

1097

786− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" the"},"finish_reason":null}]}

1098+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"fmt"},"finish_reason":null}]}

787

1099

788− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `fmt"},"finish_reason":null}]}

1100+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".Println"},"finish_reason":null}]}

789

1101

790− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".Println"},"finish_reason":null}]}

1102+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`"},"finish_reason":null}]}

791

1103

792− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"` line"},"finish_reason":null}]}

1104+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" line"},"finish_reason":null}]}

793

1105

794− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"."},"finish_reason":null}]}

1106+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"."},"finish_reason":null}]}

795

1107

796− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

1108+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

797

1109

798− data: {"id":"chatcmpl-af33a8b429bef9937266c050","object":"chat.completion.chunk","created":1779218915,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":107,"completion_tokens":33,"total_tokens":8204,"prompt_tokens_details":{"cached_tokens":8064},"completion_tokens_details":{"reasoning_tokens":32},"cost":{"usd":0.0004465,"hypercredits":0.00893},"remaining":{"hypercredits":0}}}

1110+ data: {"id":"chatcmpl-7ec589f575c15e2f5709ca1c","object":"chat.completion.chunk","created":1780069700,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":96,"completion_tokens":28,"total_tokens":8188,"prompt_tokens_details":{"cached_tokens":8064},"cost":{"usd":0.00024336000000000003,"hypercredits":0.0048672},"remaining":{"hypercredits":0}}}

799

1111

800 data: [DONE]

1112 data: [DONE]

801

1113

@@ -804,4 +1116,4 @@ interactions:

804 - text/event-stream

1116 - text/event-stream

805 status: 200 OK

1117 status: 200 OK

806 code: 200

1118 code: 200

807− duration: 1.276323125s

1119+ duration: 958.960958ms

internal/agent/testdata/TestCoderAgent/glm-5.1/parallel_tool_calls.yaml+231−375

@@ -2,64 +2,6 @@

2version: 2

3interactions:

4- id: 0

5− request:

6− proto: HTTP/1.1

7− proto_major: 1

8− proto_minor: 1

9− content_length: 34054

10− host: ""

11− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/parallel_tool_calls\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use glob to find all .go files and use ls to list the current directory, it is very important that you run both tool calls in parallel\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

12− headers:

13− Accept:

14− - application/json

15− Content-Type:

16− - application/json

17− User-Agent:

18− - Charm-Crush/devel (https://charm.land/crush)

19− url: https://hyper.charm.land/v1/chat/completions

20− method: POST

21− response:

22− proto: HTTP/2.0

23− proto_major: 2

24− proto_minor: 0

25− content_length: -1

26− uncompressed: true

27− body: |+

28− data: {"id":"chatcmpl-a277ccc142e9d09f8e315509","object":"chat.completion.chunk","created":1779218939,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

29−

30− data: {"id":"chatcmpl-a277ccc142e9d09f8e315509","object":"chat.completion.chunk","created":1779218939,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_e73938ee068547bc9b3270b8","type":"function","index":0,"function":{"name":"glob","arguments":""}}]},"finish_reason":null}]}

31−

32− data: {"id":"chatcmpl-a277ccc142e9d09f8e315509","object":"chat.completion.chunk","created":1779218939,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

33−

34− data: {"id":"chatcmpl-a277ccc142e9d09f8e315509","object":"chat.completion.chunk","created":1779218939,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

35−

36− data: {"id":"chatcmpl-a277ccc142e9d09f8e315509","object":"chat.completion.chunk","created":1779218939,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"pattern\": "}}]},"finish_reason":null}]}

37−

38− data: {"id":"chatcmpl-a277ccc142e9d09f8e315509","object":"chat.completion.chunk","created":1779218939,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"**/*."}}]},"finish_reason":null}]}

39−

40− data: {"id":"chatcmpl-a277ccc142e9d09f8e315509","object":"chat.completion.chunk","created":1779218939,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"go\""}}]},"finish_reason":null}]}

41−

42− data: {"id":"chatcmpl-a277ccc142e9d09f8e315509","object":"chat.completion.chunk","created":1779218939,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"}"}}]},"finish_reason":null}]}

43−

44− data: {"id":"chatcmpl-a277ccc142e9d09f8e315509","object":"chat.completion.chunk","created":1779218939,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_1354bab8e54b44cc82a27886","type":"function","index":1,"function":{"name":"ls","arguments":""}}]},"finish_reason":null}]}

45−

46− data: {"id":"chatcmpl-a277ccc142e9d09f8e315509","object":"chat.completion.chunk","created":1779218939,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":1,"function":{"arguments":""}}]},"finish_reason":null}]}

47−

48− data: {"id":"chatcmpl-a277ccc142e9d09f8e315509","object":"chat.completion.chunk","created":1779218939,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":1,"function":{"arguments":"{}"}}]},"finish_reason":null}]}

49−

50− data: {"id":"chatcmpl-a277ccc142e9d09f8e315509","object":"chat.completion.chunk","created":1779218939,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

51−

52− data: {"id":"chatcmpl-a277ccc142e9d09f8e315509","object":"chat.completion.chunk","created":1779218939,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":731,"completion_tokens":16,"total_tokens":7851,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":15},"cost":{"usd":0.0012329,"hypercredits":0.024658},"remaining":{"hypercredits":0}}}

53−

54− data: [DONE]

55−

56− headers:

57− Content-Type:

58− - text/event-stream

59− status: 200 OK

60− code: 200

61− duration: 1.784268666s

62−- id: 1

63 request:

5 request:

64 proto: HTTP/1.1

6 proto: HTTP/1.1

65 proto_major: 1

7 proto_major: 1

@@ -83,461 +25,355 @@ interactions:

83 content_length: -1

25 content_length: -1

84 uncompressed: true

26 uncompressed: true

85 body: |+

27 body: |+

86− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

28+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

87−

88− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

89−

90− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

91−

92− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" wants"},"finish_reason":null}]}

93−

94− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

95−

96− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

97−

98− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

99−

100− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" summar"},"finish_reason":null}]}

101−

102− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"izing"},"finish_reason":null}]}

103−

104− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

105−

106− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

107−

108− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

109−

110− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" The"},"finish_reason":null}]}

111−

112− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

113−

114− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

115−

116− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

117−

118− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"use"},"finish_reason":null}]}

119−

120− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" glob"},"finish_reason":null}]}

121−

122− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

123−

124− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" find"},"finish_reason":null}]}

125−

126− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" all"},"finish_reason":null}]}

127−

128− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

129−

130− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"go"},"finish_reason":null}]}

131−

132− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

133−

134− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

135−

136− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" use"},"finish_reason":null}]}

137−

138− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ls"},"finish_reason":null}]}

139−

140− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

141−

142− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" list"},"finish_reason":null}]}

143−

144− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

145−

146− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" current"},"finish_reason":null}]}

147−

148− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" directory"},"finish_reason":null}]}

149−

150− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

151

152− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" it"},"finish_reason":null}]}

30+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

153

154− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" is"},"finish_reason":null}]}

32+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

155

156− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" very"},"finish_reason":null}]}

34+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" asks"},"finish_reason":null}]}

157

158− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" important"},"finish_reason":null}]}

36+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

159

160− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" that"},"finish_reason":null}]}

38+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

161

162− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" you"},"finish_reason":null}]}

40+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Generate"},"finish_reason":null}]}

163

164− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" run"},"finish_reason":null}]}

42+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

165

166− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" both"},"finish_reason":null}]}

44+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

167

168− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" tool"},"finish_reason":null}]}

46+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

169

170− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" calls"},"finish_reason":null}]}

48+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

171

172− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

50+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

173

174− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" parallel"},"finish_reason":null}]}

52+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" following"},"finish_reason":null}]}

175

176− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

54+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

177

178− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Need"},"finish_reason":null}]}

56+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

179

180− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

58+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" use"},"finish_reason":null}]}

181

182− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ≤"},"finish_reason":null}]}

60+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" glob"},"finish_reason":null}]}

183

184− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

62+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

185

186− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

64+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" find"},"finish_reason":null}]}

187

188− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

66+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" all"},"finish_reason":null}]}

189

190− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

68+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

191

192− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

70+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"go"},"finish_reason":null}]}

193

194− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

72+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

195

196− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

74+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

197

198− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" colon"},"finish_reason":null}]}

76+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" use"},"finish_reason":null}]}

199

200− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

78+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ls"},"finish_reason":null}]}

201

202− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" one"},"finish_reason":null}]}

80+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

203

204− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

82+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" list"},"finish_reason":null}]}

205

206− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

84+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

207

208− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Possibly"},"finish_reason":null}]}

86+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" current"},"finish_reason":null}]}

209

210− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

88+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" directory"},"finish_reason":null}]}

211

212− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Parallel"},"finish_reason":null}]}

90+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

213

214− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" glob"},"finish_reason":null}]}

92+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" it"},"finish_reason":null}]}

215

216− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

94+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" is"},"finish_reason":null}]}

217

218− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

96+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" very"},"finish_reason":null}]}

219

220− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"go"},"finish_reason":null}]}

98+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" important"},"finish_reason":null}]}

221

222− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

100+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" that"},"finish_reason":null}]}

223

101

224− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

102+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" you"},"finish_reason":null}]}

225

103

226− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ls"},"finish_reason":null}]}

104+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" run"},"finish_reason":null}]}

227

105

228− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" current"},"finish_reason":null}]}

106+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" both"},"finish_reason":null}]}

229

107

230− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" dir"},"finish_reason":null}]}

108+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" tool"},"finish_reason":null}]}

231

109

232− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

110+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" calls"},"finish_reason":null}]}

233

111

234− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

112+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

235

113

236− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

114+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" parallel"},"finish_reason":null}]}

237

115

238− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

116+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

239

117

240− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

118+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Need"},"finish_reason":null}]}

241

119

242− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Parallel"},"finish_reason":null}]}

120+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

243

121

244− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" glob"},"finish_reason":null}]}

122+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

245

123

246− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

124+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" same"},"finish_reason":null}]}

247

125

248− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

126+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" language"},"finish_reason":null}]}

249

127

250− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"go"},"finish_reason":null}]}

128+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

251

129

252− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

130+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"English"},"finish_reason":null}]}

253

131

254− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

132+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"),"},"finish_reason":null}]}

255

133

256− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ls"},"finish_reason":null}]}

134+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ≤"},"finish_reason":null}]}

257

135

258− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" current"},"finish_reason":null}]}

136+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

259

137

260− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" dir"},"finish_reason":null}]}

138+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

261

139

262− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

140+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

263

141

264− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Let's"},"finish_reason":null}]}

142+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" one"},"finish_reason":null}]}

265

143

266− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

144+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

267

145

268− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

146+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

269

147

270− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" P"},"finish_reason":null}]}

148+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

271

149

272− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

150+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

273

151

274− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

152+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" or"},"finish_reason":null}]}

275

153

276− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

154+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" col"},"finish_reason":null}]}

277

155

278− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

156+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ons"},"finish_reason":null}]}

279

157

280− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"2"},"finish_reason":null}]}

158+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

281

159

282− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" r"},"finish_reason":null}]}

160+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Summ"},"finish_reason":null}]}

283

161

284− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"3"},"finish_reason":null}]}

162+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ar"},"finish_reason":null}]}

285

163

286− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

164+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ize"},"finish_reason":null}]}

287

165

288− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

166+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

289

167

290− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" l"},"finish_reason":null}]}

168+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

291

169

292− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"5"},"finish_reason":null}]}

170+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Parallel"},"finish_reason":null}]}

293

171

294− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" l"},"finish_reason":null}]}

172+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" glob"},"finish_reason":null}]}

295

173

296− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"6"},"finish_reason":null}]}

174+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

297

175

298− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

176+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"go"},"finish_reason":null}]}

299

177

300− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"7"},"finish_reason":null}]}

178+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

301

179

302− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" l"},"finish_reason":null}]}

180+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

303

181

304− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"8"},"finish_reason":null}]}

182+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ls"},"finish_reason":null}]}

305

183

306− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

184+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" directory"},"finish_reason":null}]}

307

185

308− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

186+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

309

187

310− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

188+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

311

189

312− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"9"},"finish_reason":null}]}

190+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

313

191

314− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" g"},"finish_reason":null}]}

192+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

315

193

316− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"10"},"finish_reason":null}]}

194+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Let's"},"finish_reason":null}]}

317

195

318− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" l"},"finish_reason":null}]}

196+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

319

197

320− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"11"},"finish_reason":null}]}

198+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

321

199

322− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" o"},"finish_reason":null}]}

200+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

323

201

324− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"12"},"finish_reason":null}]}

202+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Parallel"},"finish_reason":null}]}

325

203

326− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" b"},"finish_reason":null}]}

204+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" glob"},"finish_reason":null}]}

327

205

328− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"13"},"finish_reason":null}]}

206+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

329

207

330− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

208+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"go"},"finish_reason":null}]}

331

209

332− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

210+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

333

211

334− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

212+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

335

213

336− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"14"},"finish_reason":null}]}

214+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ls"},"finish_reason":null}]}

337

215

338− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" f"},"finish_reason":null}]}

216+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" directory"},"finish_reason":null}]}

339

217

340− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"15"},"finish_reason":null}]}

218+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

341

219

342− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" o"},"finish_reason":null}]}

220+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

343

221

344− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"16"},"finish_reason":null}]}

222+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

345

223

346− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" r"},"finish_reason":null}]}

224+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"39"},"finish_reason":null}]}

347

225

348− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"17"},"finish_reason":null}]}

226+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

349

227

350− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

228+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" including"},"finish_reason":null}]}

351

229

352− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

230+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" spaces"},"finish_reason":null}]}

353

231

354− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

232+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

355

233

356− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"18"},"finish_reason":null}]}

234+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

357

235

358− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

236+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" good"},"finish_reason":null}]}

359

237

360− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"19"},"finish_reason":null}]}

238+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

361

239

362− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" g"},"finish_reason":null}]}

240+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Or"},"finish_reason":null}]}

363

241

364− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"20"},"finish_reason":null}]}

242+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

365

243

366− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" o"},"finish_reason":null}]}

244+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Run"},"finish_reason":null}]}

367

245

368− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"21"},"finish_reason":null}]}

246+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" glob"},"finish_reason":null}]}

369

247

370− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

248+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

371

249

372− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

250+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ."},"finish_reason":null}]}

373

251

374− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

252+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"go"},"finish_reason":null}]}

375

253

376− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"22"},"finish_reason":null}]}

254+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" files"},"finish_reason":null}]}

377

255

378− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" f"},"finish_reason":null}]}

256+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

379

257

380− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"23"},"finish_reason":null}]}

258+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ls"},"finish_reason":null}]}

381

259

382− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" i"},"finish_reason":null}]}

260+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

383

261

384− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"24"},"finish_reason":null}]}

262+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" parallel"},"finish_reason":null}]}

385

263

386− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" l"},"finish_reason":null}]}

264+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

387

265

388− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"25"},"finish_reason":null}]}

266+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

389

267

390− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

268+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

391

269

392− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"26"},"finish_reason":null}]}

270+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"41"},"finish_reason":null}]}

393

271

394− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" s"},"finish_reason":null}]}

272+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

395

273

396− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"27"},"finish_reason":null}]}

274+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

397

275

398− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

276+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" No"},"finish_reason":null}]}

399

277

400− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

278+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" colon"},"finish_reason":null}]}

401

279

402− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

280+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

403

281

404− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"28"},"finish_reason":null}]}

282+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" So"},"finish_reason":null}]}

405

283

406− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

284+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" output"},"finish_reason":null}]}

407

285

408− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"29"},"finish_reason":null}]}

286+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" one"},"finish_reason":null}]}

409

287

410− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" n"},"finish_reason":null}]}

288+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

411

289

412− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"30"},"finish_reason":null}]}

290+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".\n\n"},"finish_reason":null}]}

413

291

414− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" d"},"finish_reason":null}]}

292+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Parallel"},"finish_reason":null}]}

415

293

416− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"31"},"finish_reason":null}]}

294+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" glob"},"finish_reason":null}]}

417

295

418− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

296+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" ."},"finish_reason":null}]}

419

297

420− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

298+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"go"},"finish_reason":null}]}

421

299

422− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

300+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" files"},"finish_reason":null}]}

423

301

424− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"32"},"finish_reason":null}]}

302+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" and"},"finish_reason":null}]}

425

303

426− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" l"},"finish_reason":null}]}

304+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" ls"},"finish_reason":null}]}

427

305

428− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"33"},"finish_reason":null}]}

306+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" directory"},"finish_reason":null}]}

429

307

430− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" s"},"finish_reason":null}]}

308+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

431

309

432− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"34"},"finish_reason":null}]}

310+ data: {"id":"chatcmpl-fba321fc554468b912a037cf","object":"chat.completion.chunk","created":1780069707,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":233,"completion_tokens":149,"total_tokens":382,"completion_tokens_details":{"reasoning_tokens":132},"cost":{"usd":0.00021335,"hypercredits":0.004267},"remaining":{"hypercredits":0}}}

433

311

434− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

312+ data: [DONE]

435−

436− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

437−

438− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

439−

440− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"35"},"finish_reason":null}]}

441−

442− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" c"},"finish_reason":null}]}

443−

444− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"36"},"finish_reason":null}]}

445−

446− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" u"},"finish_reason":null}]}

447−

448− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"37"},"finish_reason":null}]}

449−

450− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" r"},"finish_reason":null}]}

451−

452− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"38"},"finish_reason":null}]}

453−

454− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" r"},"finish_reason":null}]}

455−

456− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"39"},"finish_reason":null}]}

457−

458− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

459−

460− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"40"},"finish_reason":null}]}

461−

462− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" n"},"finish_reason":null}]}

463−

464− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"41"},"finish_reason":null}]}

465−

466− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" t"},"finish_reason":null}]}

467−

468− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"42"},"finish_reason":null}]}

469−

470− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

471−

472− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

473−

474− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

475−

476− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"43"},"finish_reason":null}]}

477−

478− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" d"},"finish_reason":null}]}

479−

480− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"44"},"finish_reason":null}]}

481−

482− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" i"},"finish_reason":null}]}

483−

484− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"45"},"finish_reason":null}]}

485−

486− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" r"},"finish_reason":null}]}

487−

488− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"46"},"finish_reason":null}]}

489−

490− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ="},"finish_reason":null}]}

491−

492− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"46"},"finish_reason":null}]}

493−

494− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

495−

496− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

497−

498− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" It's"},"finish_reason":null}]}

499−

500− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" under"},"finish_reason":null}]}

501−

502− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

503−

504− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

505−

506− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

507

313

508− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Good"},"finish_reason":null}]}

314+ headers:

315+ Content-Type:

316+ - text/event-stream

317+ status: 200 OK

318+ code: 200

319+ duration: 388.11675ms

320+- id: 1

321+ request:

322+ proto: HTTP/1.1

323+ proto_major: 1

324+ proto_minor: 1

325+ content_length: 34035

326+ host: ""

327+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/parallel_tool_calls\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use glob to find all .go files and use ls to list the current directory, it is very important that you run both tool calls in parallel\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

328+ headers:

329+ Accept:

330+ - application/json

331+ Content-Type:

332+ - application/json

333+ User-Agent:

334+ - Charm-Crush/devel (https://charm.land/crush)

335+ url: https://hyper.charm.land/v1/chat/completions

336+ method: POST

337+ response:

338+ proto: HTTP/2.0

339+ proto_major: 2

340+ proto_minor: 0

341+ content_length: -1

342+ uncompressed: true

343+ body: |+

344+ data: {"id":"chatcmpl-9a13601d7adcbd5076cd2ee2","object":"chat.completion.chunk","created":1780069707,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

509

345

510− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

346+ data: {"id":"chatcmpl-9a13601d7adcbd5076cd2ee2","object":"chat.completion.chunk","created":1780069707,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_ahYKOjB3foFkbILWEFYd57Y2","type":"function","index":0,"function":{"name":"glob","arguments":""}}]},"finish_reason":null}]}

511

347

512− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Return"},"finish_reason":null}]}

348+ data: {"id":"chatcmpl-9a13601d7adcbd5076cd2ee2","object":"chat.completion.chunk","created":1780069707,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

513

349

514− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" that"},"finish_reason":null}]}

350+ data: {"id":"chatcmpl-9a13601d7adcbd5076cd2ee2","object":"chat.completion.chunk","created":1780069707,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

515

351

516− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

352+ data: {"id":"chatcmpl-9a13601d7adcbd5076cd2ee2","object":"chat.completion.chunk","created":1780069707,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"pattern\": "}}]},"finish_reason":null}]}

517

353

518− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Parallel"},"finish_reason":null}]}

354+ data: {"id":"chatcmpl-9a13601d7adcbd5076cd2ee2","object":"chat.completion.chunk","created":1780069707,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"**"}}]},"finish_reason":null}]}

519

355

520− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" glob"},"finish_reason":null}]}

356+ data: {"id":"chatcmpl-9a13601d7adcbd5076cd2ee2","object":"chat.completion.chunk","created":1780069707,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/*."}}]},"finish_reason":null}]}

521

357

522− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" for"},"finish_reason":null}]}

358+ data: {"id":"chatcmpl-9a13601d7adcbd5076cd2ee2","object":"chat.completion.chunk","created":1780069707,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"go"}}]},"finish_reason":null}]}

523

359

524− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" ."},"finish_reason":null}]}

360+ data: {"id":"chatcmpl-9a13601d7adcbd5076cd2ee2","object":"chat.completion.chunk","created":1780069707,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"}"}}]},"finish_reason":null}]}

525

361

526− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"go"},"finish_reason":null}]}

362+ data: {"id":"chatcmpl-9a13601d7adcbd5076cd2ee2","object":"chat.completion.chunk","created":1780069707,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_9FOIBVsp5OvFXp01VgCefCPx","type":"function","index":1,"function":{"name":"ls","arguments":""}}]},"finish_reason":null}]}

527

363

528− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" files"},"finish_reason":null}]}

364+ data: {"id":"chatcmpl-9a13601d7adcbd5076cd2ee2","object":"chat.completion.chunk","created":1780069707,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":1,"function":{"arguments":""}}]},"finish_reason":null}]}

529

365

530− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" and"},"finish_reason":null}]}

366+ data: {"id":"chatcmpl-9a13601d7adcbd5076cd2ee2","object":"chat.completion.chunk","created":1780069707,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":1,"function":{"arguments":"{"}}]},"finish_reason":null}]}

531

367

532− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" ls"},"finish_reason":null}]}

368+ data: {"id":"chatcmpl-9a13601d7adcbd5076cd2ee2","object":"chat.completion.chunk","created":1780069707,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":1,"function":{"arguments":"\"depth\": "}}]},"finish_reason":null}]}

533

369

534− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" current"},"finish_reason":null}]}

370+ data: {"id":"chatcmpl-9a13601d7adcbd5076cd2ee2","object":"chat.completion.chunk","created":1780069707,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":1,"function":{"arguments":"3"}}]},"finish_reason":null}]}

535

371

536− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" dir"},"finish_reason":null}]}

372+ data: {"id":"chatcmpl-9a13601d7adcbd5076cd2ee2","object":"chat.completion.chunk","created":1780069707,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":1,"function":{"arguments":"}"}}]},"finish_reason":null}]}

537

373

538− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

374+ data: {"id":"chatcmpl-9a13601d7adcbd5076cd2ee2","object":"chat.completion.chunk","created":1780069707,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

539

375

540− data: {"id":"chatcmpl-defd356e22c24a5625fb8de4","object":"chat.completion.chunk","created":1779218938,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":165,"completion_tokens":235,"total_tokens":469,"prompt_tokens_details":{"cached_tokens":69},"completion_tokens_details":{"reasoning_tokens":170},"cost":{"usd":0.0001785,"hypercredits":0.00357},"remaining":{"hypercredits":0}}}

376+ data: {"id":"chatcmpl-9a13601d7adcbd5076cd2ee2","object":"chat.completion.chunk","created":1780069707,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":726,"completion_tokens":22,"total_tokens":7852,"prompt_tokens_details":{"cached_tokens":7104},"cost":{"usd":0.0010434600000000002,"hypercredits":0.0208692},"remaining":{"hypercredits":0}}}

541

377

542 data: [DONE]

378 data: [DONE]

543

379

@@ -546,15 +382,15 @@ interactions:

546 - text/event-stream

382 - text/event-stream

547 status: 200 OK

383 status: 200 OK

548 code: 200

384 code: 200

549− duration: 885.880875ms

385+ duration: 972.3925ms

550- id: 2

386- id: 2

551 request:

387 request:

552 proto: HTTP/1.1

388 proto: HTTP/1.1

553 proto_major: 1

389 proto_major: 1

554 proto_minor: 1

390 proto_minor: 1

555− content_length: 34622

391+ content_length: 34615

556 host: ""

392 host: ""

557− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/parallel_tool_calls\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use glob to find all .go files and use ls to list the current directory, it is very important that you run both tool calls in parallel\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_e73938ee068547bc9b3270b8\",\"function\":{\"arguments\":\"{\\\"pattern\\\": \\\"**/*.go\\\"}\",\"name\":\"glob\"},\"type\":\"function\"},{\"id\":\"call_1354bab8e54b44cc82a27886\",\"function\":{\"arguments\":\"{}\",\"name\":\"ls\"},\"type\":\"function\"}],\"role\":\"assistant\"},{\"content\":\"/tmp/crush-test/TestCoderAgent/glm-5.1/parallel_tool_calls/main.go\",\"tool_call_id\":\"call_e73938ee068547bc9b3270b8\",\"role\":\"tool\"},{\"content\":\"\\n- /tmp/crush-test/TestCoderAgent/glm-5.1/parallel_tool_calls/\\n - go.mod\\n - main.go\\n\",\"tool_call_id\":\"call_1354bab8e54b44cc82a27886\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

393+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/parallel_tool_calls\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use glob to find all .go files and use ls to list the current directory, it is very important that you run both tool calls in parallel\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_ahYKOjB3foFkbILWEFYd57Y2\",\"function\":{\"arguments\":\"{\\\"pattern\\\": \\\"**/*.go\\\"}\",\"name\":\"glob\"},\"type\":\"function\"},{\"id\":\"call_9FOIBVsp5OvFXp01VgCefCPx\",\"function\":{\"arguments\":\"{\\\"depth\\\": 3}\",\"name\":\"ls\"},\"type\":\"function\"}],\"role\":\"assistant\"},{\"content\":\"/tmp/crush-test/TestCoderAgent/glm-5.1/parallel_tool_calls/main.go\",\"tool_call_id\":\"call_ahYKOjB3foFkbILWEFYd57Y2\",\"role\":\"tool\"},{\"content\":\"\\n- /tmp/crush-test/TestCoderAgent/glm-5.1/parallel_tool_calls/\\n - go.mod\\n - main.go\\n\",\"tool_call_id\":\"call_9FOIBVsp5OvFXp01VgCefCPx\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

558 headers:

394 headers:

559 Accept:

395 Accept:

560 - application/json

396 - application/json

@@ -571,39 +407,59 @@ interactions:

571 content_length: -1

407 content_length: -1

572 uncompressed: true

408 uncompressed: true

573 body: |+

409 body: |+

574− data: {"id":"chatcmpl-9cfc37acf8a2a874c7582d1b","object":"chat.completion.chunk","created":1779218941,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

410+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

411+

412+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Found"},"finish_reason":null}]}

413+

414+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" "},"finish_reason":null}]}

415+

416+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"1"},"finish_reason":null}]}

417+

418+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `."},"finish_reason":null}]}

419+

420+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"go"},"finish_reason":null}]}

421+

422+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`"},"finish_reason":null}]}

423+

424+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" file"},"finish_reason":null}]}

425+

426+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":":"},"finish_reason":null}]}

427+

428+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

429+

430+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"main"},"finish_reason":null}]}

575

431

576− data: {"id":"chatcmpl-9cfc37acf8a2a874c7582d1b","object":"chat.completion.chunk","created":1779218941,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Found"},"finish_reason":null}]}

432+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".go"},"finish_reason":null}]}

577

433

578− data: {"id":"chatcmpl-9cfc37acf8a2a874c7582d1b","object":"chat.completion.chunk","created":1779218941,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" "},"finish_reason":null}]}

434+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`."},"finish_reason":null}]}

579

435

580− data: {"id":"chatcmpl-9cfc37acf8a2a874c7582d1b","object":"chat.completion.chunk","created":1779218941,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"1 `."},"finish_reason":null}]}

436+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" Current"},"finish_reason":null}]}

581

437

582− data: {"id":"chatcmpl-9cfc37acf8a2a874c7582d1b","object":"chat.completion.chunk","created":1779218941,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"go"},"finish_reason":null}]}

438+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" directory"},"finish_reason":null}]}

583

439

584− data: {"id":"chatcmpl-9cfc37acf8a2a874c7582d1b","object":"chat.completion.chunk","created":1779218941,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"` file"},"finish_reason":null}]}

440+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" contains"},"finish_reason":null}]}

585

441

586− data: {"id":"chatcmpl-9cfc37acf8a2a874c7582d1b","object":"chat.completion.chunk","created":1779218941,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":": `"},"finish_reason":null}]}

442+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

587

443

588− data: {"id":"chatcmpl-9cfc37acf8a2a874c7582d1b","object":"chat.completion.chunk","created":1779218941,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"main.go"},"finish_reason":null}]}

444+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"go"},"finish_reason":null}]}

589

445

590− data: {"id":"chatcmpl-9cfc37acf8a2a874c7582d1b","object":"chat.completion.chunk","created":1779218941,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`. The"},"finish_reason":null}]}

446+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".mod"},"finish_reason":null}]}

591

447

592− data: {"id":"chatcmpl-9cfc37acf8a2a874c7582d1b","object":"chat.completion.chunk","created":1779218941,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" directory contains"},"finish_reason":null}]}

448+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`"},"finish_reason":null}]}

593

449

594− data: {"id":"chatcmpl-9cfc37acf8a2a874c7582d1b","object":"chat.completion.chunk","created":1779218941,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `go"},"finish_reason":null}]}

450+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" and"},"finish_reason":null}]}

595

451

596− data: {"id":"chatcmpl-9cfc37acf8a2a874c7582d1b","object":"chat.completion.chunk","created":1779218941,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".mod"},"finish_reason":null}]}

452+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

597

453

598− data: {"id":"chatcmpl-9cfc37acf8a2a874c7582d1b","object":"chat.completion.chunk","created":1779218941,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"` and"},"finish_reason":null}]}

454+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"main"},"finish_reason":null}]}

599

455

600− data: {"id":"chatcmpl-9cfc37acf8a2a874c7582d1b","object":"chat.completion.chunk","created":1779218941,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `main"},"finish_reason":null}]}

456+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".go"},"finish_reason":null}]}

601

457

602− data: {"id":"chatcmpl-9cfc37acf8a2a874c7582d1b","object":"chat.completion.chunk","created":1779218941,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".go`."},"finish_reason":null}]}

458+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`."},"finish_reason":null}]}

603

459

604− data: {"id":"chatcmpl-9cfc37acf8a2a874c7582d1b","object":"chat.completion.chunk","created":1779218941,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

460+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

605

461

606− data: {"id":"chatcmpl-9cfc37acf8a2a874c7582d1b","object":"chat.completion.chunk","created":1779218941,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":99,"completion_tokens":26,"total_tokens":7933,"prompt_tokens_details":{"cached_tokens":7808},"completion_tokens_details":{"reasoning_tokens":25},"cost":{"usd":0.0003729,"hypercredits":0.007458},"remaining":{"hypercredits":0}}}

462+ data: {"id":"chatcmpl-0cb64d48ec795818094457c9","object":"chat.completion.chunk","created":1780069708,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":100,"completion_tokens":26,"total_tokens":7934,"prompt_tokens_details":{"cached_tokens":7808},"cost":{"usd":0.0002402,"hypercredits":0.004804},"remaining":{"hypercredits":0}}}

607

463

608 data: [DONE]

464 data: [DONE]

609

465

@@ -612,4 +468,4 @@ interactions:

612 - text/event-stream

468 - text/event-stream

613 status: 200 OK

469 status: 200 OK

614 code: 200

470 code: 200

615− duration: 1.564047375s

471+ duration: 507.071084ms

internal/agent/testdata/TestCoderAgent/glm-5.1/read_a_file.yaml+201−145

@@ -25,165 +25,181 @@ interactions:

25 content_length: -1

26 uncompressed: true

27 body: |+

28− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

28+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

30− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

30+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"We"},"finish_reason":null}]}

32− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

32+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" need"},"finish_reason":null}]}

34− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" says"},"finish_reason":null}]}

34+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

36− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

36+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" produce"},"finish_reason":null}]}

38− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

38+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" short"},"finish_reason":null}]}

40− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Generate"},"finish_reason":null}]}

40+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

42− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

42+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" based"},"finish_reason":null}]}

44− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

44+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" on"},"finish_reason":null}]}

46− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

46+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user's"},"finish_reason":null}]}

48− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

48+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" first"},"finish_reason":null}]}

50− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

50+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" message"},"finish_reason":null}]}

52− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" following"},"finish_reason":null}]}

52+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

54− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

54+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" User"},"finish_reason":null}]}

56− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

56+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" wrote"},"finish_reason":null}]}

58− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Read"},"finish_reason":null}]}

58+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

60− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

60+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

62− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" go"},"finish_reason":null}]}

62+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Generate"},"finish_reason":null}]}

64− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" mod"},"finish_reason":null}]}

64+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

66− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

66+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

68− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" The"},"finish_reason":null}]}

68+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

70− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

70+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

72− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" wants"},"finish_reason":null}]}

72+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

74− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

74+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" following"},"finish_reason":null}]}

76− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

76+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

78− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

78+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":\n\n"},"finish_reason":null}]}

80− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" based"},"finish_reason":null}]}

80+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Read"},"finish_reason":null}]}

82− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" on"},"finish_reason":null}]}

82+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

84− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" that"},"finish_reason":null}]}

84+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" go"},"finish_reason":null}]}

86− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

86+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" mod"},"finish_reason":null}]}

88− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

88+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\n"},"finish_reason":null}]}

90− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" So"},"finish_reason":null}]}

90+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \u003c"},"finish_reason":null}]}

92− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

92+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"think"},"finish_reason":null}]}

94− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

94+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\u003e\n\n"},"finish_reason":null}]}

96− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

96+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\u003c/"},"finish_reason":null}]}

98− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Reading"},"finish_reason":null}]}

98+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"think"},"finish_reason":null}]}

100− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

100+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\u003e\""},"finish_reason":null}]}

101

102− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" go"},"finish_reason":null}]}

102+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" This"},"finish_reason":null}]}

103

104− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".mod"},"finish_reason":null}]}

104+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" is"},"finish_reason":null}]}

105

106− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" File"},"finish_reason":null}]}

106+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" English"},"finish_reason":null}]}

107

108− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

108+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

109

110− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Keep"},"finish_reason":null}]}

110+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Title"},"finish_reason":null}]}

111

112− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" same"},"finish_reason":null}]}

112+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" summar"},"finish_reason":null}]}

113

114− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" language"},"finish_reason":null}]}

114+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"izing"},"finish_reason":null}]}

115

116− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

116+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

117

118− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" English"},"finish_reason":null}]}

118+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" maybe"},"finish_reason":null}]}

119

120− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

120+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

121

122− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Under"},"finish_reason":null}]}

122+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Reading"},"finish_reason":null}]}

123

124− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

124+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

125

126− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

126+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" go"},"finish_reason":null}]}

127

128− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

128+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".mod"},"finish_reason":null}]}

129

130− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

130+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" File"},"finish_reason":null}]}

131

132− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" No"},"finish_reason":null}]}

132+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

133

134− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

134+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Must"},"finish_reason":null}]}

135

136− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" or"},"finish_reason":null}]}

136+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

137

138− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" col"},"finish_reason":null}]}

138+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \u003c="},"finish_reason":null}]}

139

140− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ons"},"finish_reason":null}]}

140+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

141

142− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

142+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

143

144− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" One"},"finish_reason":null}]}

144+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

145

146− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

146+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" one"},"finish_reason":null}]}

147

148− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

148+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

149

150− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Should"},"finish_reason":null}]}

150+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

151

152− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

152+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

153

154− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

154+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

155

156− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Reading"},"finish_reason":null}]}

156+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"/"},"finish_reason":null}]}

157

158− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

158+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"col"},"finish_reason":null}]}

159

160− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" go"},"finish_reason":null}]}

160+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ons"},"finish_reason":null}]}

161

162− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".mod"},"finish_reason":null}]}

162+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

163

164− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" file"},"finish_reason":null}]}

164+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" So"},"finish_reason":null}]}

165

166− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

166+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" output"},"finish_reason":null}]}

167

168− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

168+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

169

170− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" fine"},"finish_reason":null}]}

170+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

171

172− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

172+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Reading"},"finish_reason":null}]}

173

174− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Reading"},"finish_reason":null}]}

174+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

175

176− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" the"},"finish_reason":null}]}

176+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" go"},"finish_reason":null}]}

177

178− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" go"},"finish_reason":null}]}

178+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".mod"},"finish_reason":null}]}

179

180− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":".mod"},"finish_reason":null}]}

180+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" file"},"finish_reason":null}]}

181

182− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" file"},"finish_reason":null}]}

182+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

183

184− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

184+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

185

186− data: {"id":"chatcmpl-6bc8cbaf35208ee8e4513d36","object":"chat.completion.chunk","created":1779218850,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":192,"completion_tokens":87,"total_tokens":279,"completion_tokens_details":{"reasoning_tokens":77},"cost":{"usd":0.0001,"hypercredits":0.002},"remaining":{"hypercredits":0}}}

186+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" fine"},"finish_reason":null}]}

187+

188+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

189+

190+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Reading"},"finish_reason":null}]}

191+

192+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" the"},"finish_reason":null}]}

193+

194+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" go"},"finish_reason":null}]}

195+

196+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":".mod"},"finish_reason":null}]}

197+

198+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" file"},"finish_reason":null}]}

199+

200+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

201+

202+ data: {"id":"chatcmpl-b7a3d84dcc5dde860a80f343","object":"chat.completion.chunk","created":1780069669,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":208,"completion_tokens":95,"total_tokens":303,"completion_tokens_details":{"reasoning_tokens":81},"cost":{"usd":0.0001448,"hypercredits":0.0028959999999999997},"remaining":{"hypercredits":0}}}

187

203

188 data: [DONE]

204 data: [DONE]

189

205

@@ -192,15 +208,15 @@ interactions:

192 - text/event-stream

208 - text/event-stream

193 status: 200 OK

209 status: 200 OK

194 code: 200

210 code: 200

195− duration: 641.109792ms

211+ duration: 394.481292ms

196- id: 1

212- id: 1

197 request:

213 request:

198 proto: HTTP/1.1

214 proto: HTTP/1.1

199 proto_major: 1

215 proto_major: 1

200 proto_minor: 1

216 proto_minor: 1

201− content_length: 33927

217+ content_length: 33908

202 host: ""

218 host: ""

203− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/read_a_file\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"Read the go mod\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

219+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/read_a_file\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"Read the go mod\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

204 headers:

220 headers:

205 Accept:

221 Accept:

206 - application/json

222 - application/json

@@ -217,41 +233,65 @@ interactions:

217 content_length: -1

233 content_length: -1

218 uncompressed: true

234 uncompressed: true

219 body: |+

235 body: |+

220− data: {"id":"chatcmpl-67276b41326a988beb00e348","object":"chat.completion.chunk","created":1779218852,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

236+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

237+

238+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

239+

240+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

241+

242+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" wants"},"finish_reason":null}]}

243+

244+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" me"},"finish_reason":null}]}

245+

246+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

247+

248+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" read"},"finish_reason":null}]}

221

249

222− data: {"id":"chatcmpl-67276b41326a988beb00e348","object":"chat.completion.chunk","created":1779218852,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

250+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

223

251

224− data: {"id":"chatcmpl-67276b41326a988beb00e348","object":"chat.completion.chunk","created":1779218852,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

252+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" go"},"finish_reason":null}]}

225

253

226− data: {"id":"chatcmpl-67276b41326a988beb00e348","object":"chat.completion.chunk","created":1779218852,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" wants me"},"finish_reason":null}]}

254+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":".mod"},"finish_reason":null}]}

227

255

228− data: {"id":"chatcmpl-67276b41326a988beb00e348","object":"chat.completion.chunk","created":1779218852,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" to read a"},"finish_reason":null}]}

256+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" file"},"finish_reason":null}]}

229

257

230− data: {"id":"chatcmpl-67276b41326a988beb00e348","object":"chat.completion.chunk","created":1779218852,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" go.mod"},"finish_reason":null}]}

258+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

231

259

232− data: {"id":"chatcmpl-67276b41326a988beb00e348","object":"chat.completion.chunk","created":1779218852,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" file. Let"},"finish_reason":null}]}

260+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" Let"},"finish_reason":null}]}

233

261

234− data: {"id":"chatcmpl-67276b41326a988beb00e348","object":"chat.completion.chunk","created":1779218852,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" me find it"},"finish_reason":null}]}

262+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" me"},"finish_reason":null}]}

235

263

236− data: {"id":"chatcmpl-67276b41326a988beb00e348","object":"chat.completion.chunk","created":1779218852,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" first."},"finish_reason":null}]}

264+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" find"},"finish_reason":null}]}

237

265

238− data: {"id":"chatcmpl-67276b41326a988beb00e348","object":"chat.completion.chunk","created":1779218852,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_604d7cc1de8040be97f5fff3","type":"function","index":0,"function":{"name":"glob","arguments":""}}]},"finish_reason":null}]}

266+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

239

267

240− data: {"id":"chatcmpl-67276b41326a988beb00e348","object":"chat.completion.chunk","created":1779218852,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

268+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" read"},"finish_reason":null}]}

241

269

242− data: {"id":"chatcmpl-67276b41326a988beb00e348","object":"chat.completion.chunk","created":1779218852,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

270+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" it"},"finish_reason":null}]}

243

271

244− data: {"id":"chatcmpl-67276b41326a988beb00e348","object":"chat.completion.chunk","created":1779218852,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"pattern\": "}}]},"finish_reason":null}]}

272+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

245

273

246− data: {"id":"chatcmpl-67276b41326a988beb00e348","object":"chat.completion.chunk","created":1779218852,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"**/"}}]},"finish_reason":null}]}

274+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_7Nc5SiqEXN8EoyKZOuMcS6tr","type":"function","index":0,"function":{"name":"glob","arguments":""}}]},"finish_reason":null}]}

247

275

248− data: {"id":"chatcmpl-67276b41326a988beb00e348","object":"chat.completion.chunk","created":1779218852,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"go.mod\""}}]},"finish_reason":null}]}

276+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

249

277

250− data: {"id":"chatcmpl-67276b41326a988beb00e348","object":"chat.completion.chunk","created":1779218852,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"}"}}]},"finish_reason":null}]}

278+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

251

279

252− data: {"id":"chatcmpl-67276b41326a988beb00e348","object":"chat.completion.chunk","created":1779218852,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

280+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"pattern\": "}}]},"finish_reason":null}]}

253

281

254− data: {"id":"chatcmpl-67276b41326a988beb00e348","object":"chat.completion.chunk","created":1779218852,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":705,"completion_tokens":31,"total_tokens":7840,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":17},"cost":{"usd":0.0012687000000000002,"hypercredits":0.025374000000000004},"remaining":{"hypercredits":0}}}

282+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"**"}}]},"finish_reason":null}]}

283+

284+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/"}}]},"finish_reason":null}]}

285+

286+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"go"}}]},"finish_reason":null}]}

287+

288+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".mod"}}]},"finish_reason":null}]}

289+

290+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"}"}}]},"finish_reason":null}]}

291+

292+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

293+

294+ data: {"id":"chatcmpl-6c1a627e1aeebb81fed69cbf","object":"chat.completion.chunk","created":1780069671,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":700,"completion_tokens":32,"total_tokens":7836,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":17},"cost":{"usd":0.0011228000000000002,"hypercredits":0.022456000000000004},"remaining":{"hypercredits":0}}}

255

295

256 data: [DONE]

296 data: [DONE]

257

297

@@ -260,15 +300,15 @@ interactions:

260 - text/event-stream

300 - text/event-stream

261 status: 200 OK

301 status: 200 OK

262 code: 200

302 code: 200

263− duration: 2.4505925s

303+ duration: 1.828266958s

264- id: 2

304- id: 2

265 request:

305 request:

266 proto: HTTP/1.1

306 proto: HTTP/1.1

267 proto_major: 1

307 proto_major: 1

268 proto_minor: 1

308 proto_minor: 1

269− content_length: 34308

309+ content_length: 34294

270 host: ""

310 host: ""

271− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/read_a_file\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"Read the go mod\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_604d7cc1de8040be97f5fff3\",\"function\":{\"arguments\":\"{\\\"pattern\\\": \\\"**/go.mod\\\"}\",\"name\":\"glob\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"The user wants me to read a go.mod file. Let me find it first.\"},{\"content\":\"/tmp/crush-test/TestCoderAgent/glm-5.1/read_a_file/go.mod\",\"tool_call_id\":\"call_604d7cc1de8040be97f5fff3\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

311+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/read_a_file\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"Read the go mod\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_7Nc5SiqEXN8EoyKZOuMcS6tr\",\"function\":{\"arguments\":\"{\\\"pattern\\\": \\\"**/go.mod\\\"}\",\"name\":\"glob\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"The user wants me to read the go.mod file. Let me find and read it.\"},{\"content\":\"/tmp/crush-test/TestCoderAgent/glm-5.1/read_a_file/go.mod\",\"tool_call_id\":\"call_7Nc5SiqEXN8EoyKZOuMcS6tr\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

272 headers:

312 headers:

273 Accept:

313 Accept:

274 - application/json

314 - application/json

@@ -285,43 +325,57 @@ interactions:

285 content_length: -1

325 content_length: -1

286 uncompressed: true

326 uncompressed: true

287 body: |+

327 body: |+

288− data: {"id":"chatcmpl-116823d7febd9686efd9217a","object":"chat.completion.chunk","created":1779218855,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

328+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

329+

330+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_eiYecsjzoLpukgMCT8EogCZD","type":"function","index":0,"function":{"name":"view","arguments":""}}]},"finish_reason":null}]}

289

331

290− data: {"id":"chatcmpl-116823d7febd9686efd9217a","object":"chat.completion.chunk","created":1779218855,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_1f8fde70bba04f24a2ef6ac1","type":"function","index":0,"function":{"name":"view","arguments":""}}]},"finish_reason":null}]}

332+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

291

333

292− data: {"id":"chatcmpl-116823d7febd9686efd9217a","object":"chat.completion.chunk","created":1779218855,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

334+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

293

335

294− data: {"id":"chatcmpl-116823d7febd9686efd9217a","object":"chat.completion.chunk","created":1779218855,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

336+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"file_path\": "}}]},"finish_reason":null}]}

295

337

296− data: {"id":"chatcmpl-116823d7febd9686efd9217a","object":"chat.completion.chunk","created":1779218855,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"file_path\": "}}]},"finish_reason":null}]}

338+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"/tmp"}}]},"finish_reason":null}]}

297

339

298− data: {"id":"chatcmpl-116823d7febd9686efd9217a","object":"chat.completion.chunk","created":1779218855,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"/tmp/c"}}]},"finish_reason":null}]}

340+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/c"}}]},"finish_reason":null}]}

299

341

300− data: {"id":"chatcmpl-116823d7febd9686efd9217a","object":"chat.completion.chunk","created":1779218855,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"rush"}}]},"finish_reason":null}]}

342+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"rush"}}]},"finish_reason":null}]}

301

343

302− data: {"id":"chatcmpl-116823d7febd9686efd9217a","object":"chat.completion.chunk","created":1779218855,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-test/Test"}}]},"finish_reason":null}]}

344+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-test"}}]},"finish_reason":null}]}

303

345

304− data: {"id":"chatcmpl-116823d7febd9686efd9217a","object":"chat.completion.chunk","created":1779218855,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Coder"}}]},"finish_reason":null}]}

346+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/Test"}}]},"finish_reason":null}]}

305

347

306− data: {"id":"chatcmpl-116823d7febd9686efd9217a","object":"chat.completion.chunk","created":1779218855,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Agent/g"}}]},"finish_reason":null}]}

348+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Coder"}}]},"finish_reason":null}]}

307

349

308− data: {"id":"chatcmpl-116823d7febd9686efd9217a","object":"chat.completion.chunk","created":1779218855,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"lm-"}}]},"finish_reason":null}]}

350+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Agent"}}]},"finish_reason":null}]}

309

351

310− data: {"id":"chatcmpl-116823d7febd9686efd9217a","object":"chat.completion.chunk","created":1779218855,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"5.1"}}]},"finish_reason":null}]}

352+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/g"}}]},"finish_reason":null}]}

311

353

312− data: {"id":"chatcmpl-116823d7febd9686efd9217a","object":"chat.completion.chunk","created":1779218855,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/read_a"}}]},"finish_reason":null}]}

354+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"lm"}}]},"finish_reason":null}]}

313

355

314− data: {"id":"chatcmpl-116823d7febd9686efd9217a","object":"chat.completion.chunk","created":1779218855,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_file"}}]},"finish_reason":null}]}

356+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-"}}]},"finish_reason":null}]}

315

357

316− data: {"id":"chatcmpl-116823d7febd9686efd9217a","object":"chat.completion.chunk","created":1779218855,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/go.mod"}}]},"finish_reason":null}]}

358+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"5"}}]},"finish_reason":null}]}

317

359

318− data: {"id":"chatcmpl-116823d7febd9686efd9217a","object":"chat.completion.chunk","created":1779218855,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\""}}]},"finish_reason":null}]}

360+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"."}}]},"finish_reason":null}]}

319

361

320− data: {"id":"chatcmpl-116823d7febd9686efd9217a","object":"chat.completion.chunk","created":1779218855,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"}"}}]},"finish_reason":null}]}

362+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"1"}}]},"finish_reason":null}]}

321

363

322− data: {"id":"chatcmpl-116823d7febd9686efd9217a","object":"chat.completion.chunk","created":1779218855,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

364+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/read"}}]},"finish_reason":null}]}

323

365

324− data: {"id":"chatcmpl-116823d7febd9686efd9217a","object":"chat.completion.chunk","created":1779218855,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":37,"completion_tokens":29,"total_tokens":7874,"prompt_tokens_details":{"cached_tokens":7808},"completion_tokens_details":{"reasoning_tokens":28},"cost":{"usd":0.0003063,"hypercredits":0.006126},"remaining":{"hypercredits":0}}}

366+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_a"}}]},"finish_reason":null}]}

367+

368+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_file"}}]},"finish_reason":null}]}

369+

370+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/go"}}]},"finish_reason":null}]}

371+

372+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".mod"}}]},"finish_reason":null}]}

373+

374+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"}"}}]},"finish_reason":null}]}

375+

376+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

377+

378+ data: {"id":"chatcmpl-cdcb80e89d80fae8a62f3656","object":"chat.completion.chunk","created":1780069672,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":96,"completion_tokens":29,"total_tokens":7869,"prompt_tokens_details":{"cached_tokens":7744},"cost":{"usd":0.00024756,"hypercredits":0.0049512},"remaining":{"hypercredits":0}}}

325

379

326 data: [DONE]

380 data: [DONE]

327

381

@@ -330,15 +384,15 @@ interactions:

330 - text/event-stream

384 - text/event-stream

331 status: 200 OK

385 status: 200 OK

332 code: 200

386 code: 200

333− duration: 2.617324083s

387+ duration: 1.298647583s

334- id: 3

388- id: 3

335 request:

389 request:

336 proto: HTTP/1.1

390 proto: HTTP/1.1

337 proto_major: 1

391 proto_major: 1

338 proto_minor: 1

392 proto_minor: 1

339− content_length: 34710

393+ content_length: 34696

340 host: ""

394 host: ""

341− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/read_a_file\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"Read the go mod\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_604d7cc1de8040be97f5fff3\",\"function\":{\"arguments\":\"{\\\"pattern\\\": \\\"**/go.mod\\\"}\",\"name\":\"glob\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"The user wants me to read a go.mod file. Let me find it first.\"},{\"content\":\"/tmp/crush-test/TestCoderAgent/glm-5.1/read_a_file/go.mod\",\"tool_call_id\":\"call_604d7cc1de8040be97f5fff3\",\"role\":\"tool\"},{\"tool_calls\":[{\"id\":\"call_1f8fde70bba04f24a2ef6ac1\",\"function\":{\"arguments\":\"{\\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/read_a_file/go.mod\\\"}\",\"name\":\"view\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"\"},{\"content\":\"<file>\\n 1|module example.com/testproject\\n 2|\\n 3|go 1.23\\n 4|\\n</file>\\n\",\"tool_call_id\":\"call_1f8fde70bba04f24a2ef6ac1\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

395+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/read_a_file\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"Read the go mod\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_7Nc5SiqEXN8EoyKZOuMcS6tr\",\"function\":{\"arguments\":\"{\\\"pattern\\\": \\\"**/go.mod\\\"}\",\"name\":\"glob\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"The user wants me to read the go.mod file. Let me find and read it.\"},{\"content\":\"/tmp/crush-test/TestCoderAgent/glm-5.1/read_a_file/go.mod\",\"tool_call_id\":\"call_7Nc5SiqEXN8EoyKZOuMcS6tr\",\"role\":\"tool\"},{\"tool_calls\":[{\"id\":\"call_eiYecsjzoLpukgMCT8EogCZD\",\"function\":{\"arguments\":\"{\\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/read_a_file/go.mod\\\"}\",\"name\":\"view\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"\"},{\"content\":\"<file>\\n 1|module example.com/testproject\\n 2|\\n 3|go 1.23\\n 4|\\n</file>\\n\",\"tool_call_id\":\"call_eiYecsjzoLpukgMCT8EogCZD\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

342 headers:

396 headers:

343 Accept:

397 Accept:

344 - application/json

398 - application/json

@@ -355,41 +409,43 @@ interactions:

355 content_length: -1

409 content_length: -1

356 uncompressed: true

410 uncompressed: true

357 body: |+

411 body: |+

358− data: {"id":"chatcmpl-df8f958cc34812301d1fd2c1","object":"chat.completion.chunk","created":1779218858,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

412+ data: {"id":"chatcmpl-0d1735ba3b3dc4655ba61003","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

413+

414+ data: {"id":"chatcmpl-0d1735ba3b3dc4655ba61003","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Module"},"finish_reason":null}]}

359

415

360− data: {"id":"chatcmpl-df8f958cc34812301d1fd2c1","object":"chat.completion.chunk","created":1779218858,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"The"},"finish_reason":null}]}

416+ data: {"id":"chatcmpl-0d1735ba3b3dc4655ba61003","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

361

417

362− data: {"id":"chatcmpl-df8f958cc34812301d1fd2c1","object":"chat.completion.chunk","created":1779218858,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

418+ data: {"id":"chatcmpl-0d1735ba3b3dc4655ba61003","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"example"},"finish_reason":null}]}

363

419

364− data: {"id":"chatcmpl-df8f958cc34812301d1fd2c1","object":"chat.completion.chunk","created":1779218858,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"go"},"finish_reason":null}]}

420+ data: {"id":"chatcmpl-0d1735ba3b3dc4655ba61003","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".com"},"finish_reason":null}]}

365

421

366− data: {"id":"chatcmpl-df8f958cc34812301d1fd2c1","object":"chat.completion.chunk","created":1779218858,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".mod` defines"},"finish_reason":null}]}

422+ data: {"id":"chatcmpl-0d1735ba3b3dc4655ba61003","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"/test"},"finish_reason":null}]}

367

423

368− data: {"id":"chatcmpl-df8f958cc34812301d1fd2c1","object":"chat.completion.chunk","created":1779218858,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":":\n\n-"},"finish_reason":null}]}

424+ data: {"id":"chatcmpl-0d1735ba3b3dc4655ba61003","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"project"},"finish_reason":null}]}

369

425

370− data: {"id":"chatcmpl-df8f958cc34812301d1fd2c1","object":"chat.completion.chunk","created":1779218858,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" **Module"},"finish_reason":null}]}

426+ data: {"id":"chatcmpl-0d1735ba3b3dc4655ba61003","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`,"},"finish_reason":null}]}

371

427

372− data: {"id":"chatcmpl-df8f958cc34812301d1fd2c1","object":"chat.completion.chunk","created":1779218858,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"**: `example"},"finish_reason":null}]}

428+ data: {"id":"chatcmpl-0d1735ba3b3dc4655ba61003","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" Go"},"finish_reason":null}]}

373

429

374− data: {"id":"chatcmpl-df8f958cc34812301d1fd2c1","object":"chat.completion.chunk","created":1779218858,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".com/test"},"finish_reason":null}]}

430+ data: {"id":"chatcmpl-0d1735ba3b3dc4655ba61003","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" "},"finish_reason":null}]}

375

431

376− data: {"id":"chatcmpl-df8f958cc34812301d1fd2c1","object":"chat.completion.chunk","created":1779218858,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"project"},"finish_reason":null}]}

432+ data: {"id":"chatcmpl-0d1735ba3b3dc4655ba61003","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"1"},"finish_reason":null}]}

377

433

378− data: {"id":"chatcmpl-df8f958cc34812301d1fd2c1","object":"chat.completion.chunk","created":1779218858,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`\n- **"},"finish_reason":null}]}

434+ data: {"id":"chatcmpl-0d1735ba3b3dc4655ba61003","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"."},"finish_reason":null}]}

379

435

380− data: {"id":"chatcmpl-df8f958cc34812301d1fd2c1","object":"chat.completion.chunk","created":1779218858,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Go version"},"finish_reason":null}]}

436+ data: {"id":"chatcmpl-0d1735ba3b3dc4655ba61003","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"23"},"finish_reason":null}]}

381

437

382− data: {"id":"chatcmpl-df8f958cc34812301d1fd2c1","object":"chat.completion.chunk","created":1779218858,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"**: 1"},"finish_reason":null}]}

438+ data: {"id":"chatcmpl-0d1735ba3b3dc4655ba61003","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":","},"finish_reason":null}]}

383

439

384− data: {"id":"chatcmpl-df8f958cc34812301d1fd2c1","object":"chat.completion.chunk","created":1779218858,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".23"},"finish_reason":null}]}

440+ data: {"id":"chatcmpl-0d1735ba3b3dc4655ba61003","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" no"},"finish_reason":null}]}

385

441

386− data: {"id":"chatcmpl-df8f958cc34812301d1fd2c1","object":"chat.completion.chunk","created":1779218858,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"\n- No"},"finish_reason":null}]}

442+ data: {"id":"chatcmpl-0d1735ba3b3dc4655ba61003","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" dependencies"},"finish_reason":null}]}

387

443

388− data: {"id":"chatcmpl-df8f958cc34812301d1fd2c1","object":"chat.completion.chunk","created":1779218858,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" dependencies"},"finish_reason":null}]}

444+ data: {"id":"chatcmpl-0d1735ba3b3dc4655ba61003","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"."},"finish_reason":null}]}

389

445

390− data: {"id":"chatcmpl-df8f958cc34812301d1fd2c1","object":"chat.completion.chunk","created":1779218858,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

446+ data: {"id":"chatcmpl-0d1735ba3b3dc4655ba61003","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

391

447

392− data: {"id":"chatcmpl-df8f958cc34812301d1fd2c1","object":"chat.completion.chunk","created":1779218858,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":104,"completion_tokens":32,"total_tokens":7944,"prompt_tokens_details":{"cached_tokens":7808},"completion_tokens_details":{"reasoning_tokens":31},"cost":{"usd":0.0004332,"hypercredits":0.008664},"remaining":{"hypercredits":0}}}

448+ data: {"id":"chatcmpl-0d1735ba3b3dc4655ba61003","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":99,"completion_tokens":18,"total_tokens":7925,"prompt_tokens_details":{"cached_tokens":7808},"cost":{"usd":0.00020529,"hypercredits":0.0041058},"remaining":{"hypercredits":0}}}

393

449

394 data: [DONE]

450 data: [DONE]

395

451

@@ -398,4 +454,4 @@ interactions:

398 - text/event-stream

454 - text/event-stream

399 status: 200 OK

455 status: 200 OK

400 code: 200

456 code: 200

401− duration: 2.357300458s

457+ duration: 998.653042ms

internal/agent/testdata/TestCoderAgent/glm-5.1/simple_test.yaml+154−74

@@ -25,135 +25,179 @@ interactions:

25 content_length: -1

26 uncompressed: true

27 body: |+

28− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

28+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

30− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"We"},"finish_reason":null}]}

30+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

32− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" need"},"finish_reason":null}]}

32+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

34− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

34+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" asks"},"finish_reason":null}]}

36− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" generate"},"finish_reason":null}]}

36+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

38− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

38+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

40− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" short"},"finish_reason":null}]}

40+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Generate"},"finish_reason":null}]}

42− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

42+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

44− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" based"},"finish_reason":null}]}

44+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

46− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" on"},"finish_reason":null}]}

46+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

48− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user's"},"finish_reason":null}]}

48+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

50− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" first"},"finish_reason":null}]}

50+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

52− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" message"},"finish_reason":null}]}

52+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" following"},"finish_reason":null}]}

54− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

54+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

56− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" The"},"finish_reason":null}]}

56+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

58− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

58+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Hello"},"finish_reason":null}]}

60− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" says"},"finish_reason":null}]}

60+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

62− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

62+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" They"},"finish_reason":null}]}

64− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

64+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" want"},"finish_reason":null}]}

66− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Generate"},"finish_reason":null}]}

66+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

68− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

68+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" short"},"finish_reason":null}]}

70− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

70+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

72− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

72+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

74− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

74+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

76− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

76+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" same"},"finish_reason":null}]}

78− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" following"},"finish_reason":null}]}

78+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" language"},"finish_reason":null}]}

80− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

80+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

82− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

82+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"English"},"finish_reason":null}]}

84− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Hello"},"finish_reason":null}]}

84+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")."},"finish_reason":null}]}

86− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

86+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Should"},"finish_reason":null}]}

88− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" So"},"finish_reason":null}]}

88+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

90− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

90+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \u003c="},"finish_reason":null}]}

92− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" should"},"finish_reason":null}]}

92+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

94− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" summarize"},"finish_reason":null}]}

94+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

96− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

96+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

98− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Hello"},"finish_reason":null}]}

98+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" So"},"finish_reason":null}]}

100− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

100+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" maybe"},"finish_reason":null}]}

101

102− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Probably"},"finish_reason":null}]}

102+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

103

104− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

104+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Greeting"},"finish_reason":null}]}

105

106− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Greeting"},"finish_reason":null}]}

106+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

107

108− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

108+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Or"},"finish_reason":null}]}

109

110− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Must"},"finish_reason":null}]}

110+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

111

112− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

112+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Hello"},"finish_reason":null}]}

113

114− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" under"},"finish_reason":null}]}

114+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

115

116− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

116+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Title"},"finish_reason":null}]}

117

118− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

118+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" should"},"finish_reason":null}]}

119

120− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

120+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

121

122− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

122+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" summary"},"finish_reason":null}]}

123

124− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" one"},"finish_reason":null}]}

124+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" of"},"finish_reason":null}]}

125

126− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

126+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user's"},"finish_reason":null}]}

127

128− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

128+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" message"},"finish_reason":null}]}

129

130− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

130+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

131

132− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

132+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" The"},"finish_reason":null}]}

133

134− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"/"},"finish_reason":null}]}

134+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

135

136− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"col"},"finish_reason":null}]}

136+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" is"},"finish_reason":null}]}

137

138− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ons"},"finish_reason":null}]}

138+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" just"},"finish_reason":null}]}

139

140− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

140+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

141

142− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" So"},"finish_reason":null}]}

142+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Hello"},"finish_reason":null}]}

143

144− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" answer"},"finish_reason":null}]}

144+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

145

146− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

146+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" So"},"finish_reason":null}]}

147

148− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Greeting"},"finish_reason":null}]}

148+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

149

150− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

150+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" could"},"finish_reason":null}]}

151

152− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Greeting"},"finish_reason":null}]}

152+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

153

154− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

154+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

155

156− data: {"id":"chatcmpl-fa252952131983a254bba099","object":"chat.completion.chunk","created":1779218848,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":143,"completion_tokens":2,"total_tokens":145,"completion_tokens_details":{"reasoning_tokens":64},"cost":{"usd":0.0001,"hypercredits":0.002},"remaining":{"hypercredits":0}}}

156+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Hello"},"finish_reason":null}]}

157+

158+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

159+

160+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" No"},"finish_reason":null}]}

161+

162+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

163+

164+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

165+

166+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

167+

168+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" colon"},"finish_reason":null}]}

169+

170+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

171+

172+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" One"},"finish_reason":null}]}

173+

174+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

175+

176+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

177+

178+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Return"},"finish_reason":null}]}

179+

180+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" just"},"finish_reason":null}]}

181+

182+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

183+

184+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".\n\n"},"finish_reason":null}]}

185+

186+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Thus"},"finish_reason":null}]}

187+

188+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" answer"},"finish_reason":null}]}

189+

190+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

191+

192+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Hello"},"finish_reason":null}]}

193+

194+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

195+

196+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Hello"},"finish_reason":null}]}

197+

198+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

199+

200+ data: {"id":"chatcmpl-0fd24293c5139ce426ac87d2","object":"chat.completion.chunk","created":1780069667,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":205,"completion_tokens":94,"total_tokens":299,"completion_tokens_details":{"reasoning_tokens":84},"cost":{"usd":0.00014548,"hypercredits":0.0029095999999999996},"remaining":{"hypercredits":0}}}

157

201

158 data: [DONE]

202 data: [DONE]

159

203

@@ -162,15 +206,15 @@ interactions:

162 - text/event-stream

206 - text/event-stream

163 status: 200 OK

207 status: 200 OK

164 code: 200

208 code: 200

165− duration: 1.428023542s

209+ duration: 965.865833ms

166- id: 1

210- id: 1

167 request:

211 request:

168 proto: HTTP/1.1

212 proto: HTTP/1.1

169 proto_major: 1

213 proto_major: 1

170 proto_minor: 1

214 proto_minor: 1

171− content_length: 33917

215+ content_length: 33898

172 host: ""

216 host: ""

173− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/simple_test\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"Hello\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

217+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/simple_test\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"Hello\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

174 headers:

218 headers:

175 Accept:

219 Accept:

176 - application/json

220 - application/json

@@ -187,15 +231,51 @@ interactions:

187 content_length: -1

231 content_length: -1

188 uncompressed: true

232 uncompressed: true

189 body: |+

233 body: |+

190− data: {"id":"chatcmpl-c24b55ff7bc482fb2d85fb0b","object":"chat.completion.chunk","created":1779218849,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

234+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

235+

236+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

237+

238+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

239+

240+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" said"},"finish_reason":null}]}

241+

242+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

243+

244+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"Hello"},"finish_reason":null}]}

245+

246+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

247+

248+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" I"},"finish_reason":null}]}

249+

250+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" should"},"finish_reason":null}]}

251+

252+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" respond"},"finish_reason":null}]}

253+

254+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" conc"},"finish_reason":null}]}

255+

256+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"is"},"finish_reason":null}]}

257+

258+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"ely"},"finish_reason":null}]}

259+

260+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

261+

262+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Hi"},"finish_reason":null}]}

263+

264+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"!"},"finish_reason":null}]}

265+

266+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" How"},"finish_reason":null}]}

267+

268+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" can"},"finish_reason":null}]}

269+

270+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" I"},"finish_reason":null}]}

191

271

192− data: {"id":"chatcmpl-c24b55ff7bc482fb2d85fb0b","object":"chat.completion.chunk","created":1779218849,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Hello"},"finish_reason":null}]}

272+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" help"},"finish_reason":null}]}

193

273

194− data: {"id":"chatcmpl-c24b55ff7bc482fb2d85fb0b","object":"chat.completion.chunk","created":1779218849,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"!"},"finish_reason":null}]}

274+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"?"},"finish_reason":null}]}

195

275

196− data: {"id":"chatcmpl-c24b55ff7bc482fb2d85fb0b","object":"chat.completion.chunk","created":1779218849,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

276+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

197

277

198− data: {"id":"chatcmpl-c24b55ff7bc482fb2d85fb0b","object":"chat.completion.chunk","created":1779218849,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":7805,"completion_tokens":4,"total_tokens":7809,"completion_tokens_details":{"reasoning_tokens":3},"cost":{"usd":0.011738299999999998,"hypercredits":0.23476599999999995},"remaining":{"hypercredits":0}}}

278+ data: {"id":"chatcmpl-879afd4304b57ccc206c58da","object":"chat.completion.chunk","created":1780069669,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":7799,"completion_tokens":22,"total_tokens":7821,"completion_tokens_details":{"reasoning_tokens":13},"cost":{"usd":0.01036369,"hypercredits":0.20727379999999998},"remaining":{"hypercredits":0}}}

199

279

200 data: [DONE]

280 data: [DONE]

201

281

@@ -204,4 +284,4 @@ interactions:

204 - text/event-stream

284 - text/event-stream

205 status: 200 OK

285 status: 200 OK

206 code: 200

286 code: 200

207− duration: 2.75824775s

287+ duration: 2.265527042s

internal/agent/testdata/TestCoderAgent/glm-5.1/sourcegraph_tool.yaml+448−458

@@ -6,9 +6,9 @@ interactions:

6 proto: HTTP/1.1

7 proto_major: 1

8 proto_minor: 1

9− content_length: 33977

9+ content_length: 819

10 host: ""

11− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/sourcegraph_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use sourcegraph to search for 'func main' in Go repositories\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

11+ body: '{"messages":[{"content":"You will generate a short title based on the first message a user begins a conversation with.\n\n<rules>\n- Keep the title in the same language that the user wrote their message in.\n- Ensure it is not more than 50 characters long.\n- The title should be a summary of the user''s message.\n- It should be one line long.\n- Do not use quotes or colons.\n- The entire text you return will be used as the title.\n- Never return anything that is more than one sentence (one line) long.\n</rules>\n\n /no_think","role":"system"},{"content":"Generate a concise title for the following content:\n\nuse sourcegraph to search for ''func main'' in Go repositories\n <think>\n\n</think>","role":"user"}],"model":"gpt-oss-120b","max_completion_tokens":40,"stream_options":{"include_usage":true},"stream":true}'

12 headers:

13 Accept:

14 - application/json

@@ -25,817 +25,807 @@ interactions:

25 content_length: -1

26 uncompressed: true

27 body: |+

28− data: {"id":"chatcmpl-f4a68f86046255eacf34bd78","object":"chat.completion.chunk","created":1779218920,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

28+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

29−

30− data: {"id":"chatcmpl-f4a68f86046255eacf34bd78","object":"chat.completion.chunk","created":1779218920,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_0fb31da44aad4c97a3ccff8f","type":"function","index":0,"function":{"name":"sourcegraph","arguments":""}}]},"finish_reason":null}]}

32− data: {"id":"chatcmpl-f4a68f86046255eacf34bd78","object":"chat.completion.chunk","created":1779218920,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

30+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"We"},"finish_reason":null}]}

34− data: {"id":"chatcmpl-f4a68f86046255eacf34bd78","object":"chat.completion.chunk","created":1779218920,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

32+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" need"},"finish_reason":null}]}

36− data: {"id":"chatcmpl-f4a68f86046255eacf34bd78","object":"chat.completion.chunk","created":1779218920,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"count\": 10"}}]},"finish_reason":null}]}

34+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

38− data: {"id":"chatcmpl-f4a68f86046255eacf34bd78","object":"chat.completion.chunk","created":1779218920,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":", "}}]},"finish_reason":null}]}

36+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" produce"},"finish_reason":null}]}

40− data: {"id":"chatcmpl-f4a68f86046255eacf34bd78","object":"chat.completion.chunk","created":1779218920,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"query\": "}}]},"finish_reason":null}]}

38+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

42− data: {"id":"chatcmpl-f4a68f86046255eacf34bd78","object":"chat.completion.chunk","created":1779218920,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"func main"}}]},"finish_reason":null}]}

40+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" short"},"finish_reason":null}]}

44− data: {"id":"chatcmpl-f4a68f86046255eacf34bd78","object":"chat.completion.chunk","created":1779218920,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" lang:Go"}}]},"finish_reason":null}]}

42+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

46− data: {"id":"chatcmpl-f4a68f86046255eacf34bd78","object":"chat.completion.chunk","created":1779218920,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\""}}]},"finish_reason":null}]}

44+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

48− data: {"id":"chatcmpl-f4a68f86046255eacf34bd78","object":"chat.completion.chunk","created":1779218920,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"}"}}]},"finish_reason":null}]}

46+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" max"},"finish_reason":null}]}

50− data: {"id":"chatcmpl-f4a68f86046255eacf34bd78","object":"chat.completion.chunk","created":1779218920,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

48+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

52− data: {"id":"chatcmpl-f4a68f86046255eacf34bd78","object":"chat.completion.chunk","created":1779218920,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":714,"completion_tokens":22,"total_tokens":7840,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":21},"cost":{"usd":0.0012602,"hypercredits":0.025203999999999997},"remaining":{"hypercredits":0}}}

50+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

54− data: [DONE]

52+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

56− headers:

54+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

57− Content-Type:

58− - text/event-stream

59− status: 200 OK

60− code: 200

61− duration: 4.934736541s

62−- id: 1

63− request:

64− proto: HTTP/1.1

65− proto_major: 1

66− proto_minor: 1

67− content_length: 424

68− host: sourcegraph.com

69− body: '{"query":"query Search($query: String!) { search(query: $query, version: V2, patternType: keyword ) { results { matchCount, limitHit, resultCount, approximateResultCount, missing { name }, timedout { name }, indexUnavailable, results { __typename, ... on FileMatch { repository { name }, file { path, url, content }, lineMatches { preview, lineNumber, offsetAndLengths } } } } } }","variables":{"query":"func main lang:Go"}}'

70− headers:

71− Content-Type:

72− - application/json

73− User-Agent:

74− - crush/1.0

75− url: https://sourcegraph.com/.api/graphql

76− method: POST

77− response:

78− proto: HTTP/2.0

79− proto_major: 2

80− proto_minor: 0

81− content_length: -1

82− uncompressed: true

83− body: '{"data":{"search":{"results":{"matchCount":30,"limitHit":true,"resultCount":30,"approximateResultCount":"30+","missing":[],"timedout":[],"indexUnavailable":false,"results":[{"__typename":"FileMatch","repository":{"name":"github.com/jmbharathram/executeoncommand"},"file":{"path":"go/channel/main.go","url":"/r/github.com/jmbharathram/executeoncommand/-/blob/go/channel/main.go","content":"package main\n\nimport \"fmt\"\n\n// Example 0\nfunc dummy_func() {\n\tfmt.Println(\"dummy_func\")\n}\nfunc main() {\n\tgo dummy_func()\n fmt.Println(\"end of main\")\n}\n\n\n// Example 1\n\npackage main\n\nimport \"fmt\"\n\nfunc dummy_func(ch chan int) {\n // receive a value from channel\n val := \u003c-ch\n\tfmt.Println(val)\n}\nfunc main() {\n // create a channel\n\tch := make(chan int)\n // trigger a parallel go routine\n\tgo dummy_func(ch)\n // send a value into the channel\n\tch \u003c- 10\n}\n\n// Example 2\n\npackage main\n\nimport (\n \"fmt\"\n //\"flag\"\n)\n\nfunc dummy_func(ch chan int) {\n // receive a value from channel\n val := \u003c-ch\n\tfmt.Println(val)\n\n //send a value again\n ch \u003c- val*2\n}\nfunc main() {\n fmt.Println(\"Welcome to main function\")\n \n //flag statements for user input\n var input_value int\n input_value = 10\n //flag.IntVar(\u0026input_value, \"v\", 10, \"Input value to send/receive via channel.\" )\n //flag.Parse()\n\n // create a channel\n\tch := make(chan int)\n\n // trigger a parallel go routine\n\tgo dummy_func(ch)\n \n // send a value into the channel\n\tch \u003c- input_value\n\n // receive a value from channel\n response := \u003c- ch\n fmt.Println(response)\n\n}\n\n// example 3\n\n\npackage main\n\nimport(\n\"fmt\"\n\"time\"\n\"math/rand\"\n)\n\nfunc routine1(ch chan int, min int, max int) {\n\n rand.Seed(time.Now().UnixNano())\n t := rand.Intn(max - min) + min\n fmt.Println(\"routine 1 is waiting: \", t, \"secs\")\n time.Sleep(time.Duration(t)*time.Second)\n\n ch \u003c- t\n}\n\nfunc routine2(ch chan int, min int, max int) {\n\n rand.Seed(time.Now().UnixNano())\n t := rand.Intn(max - min) + min\n fmt.Println(\"routine 2 is waiting: \", t, \"secs\")\n time.Sleep(time.Duration(t)*time.Second)\n\n ch \u003c- t\n}\n\nfunc main(){\n\nmin := 1\nmax := 10\n\nch1:= make(chan int)\nch2:= make(chan int)\n\ngo routine1(ch1, min, max)\ngo routine2(ch2, min, max)\n\nselect{\n\ncase \u003c- ch1:\n fmt.Println(\"routine 1 finished\")\n\ncase \u003c- ch2:\n fmt.Println(\"routine 2 finished\")\n}\n\nfmt.Println(\"end of main function\")\n\n}\n"},"lineMatches":[{"preview":"func main(){","lineNumber":106,"offsetAndLengths":[[0,9]]},{"preview":"func dummy_func() {","lineNumber":5,"offsetAndLengths":[[0,4],[11,4]]},{"preview":"\tfmt.Println(\"dummy_func\")","lineNumber":6,"offsetAndLengths":[[20,4]]},{"preview":"func main() {","lineNumber":8,"offsetAndLengths":[[0,9]]},{"preview":"\tgo dummy_func()","lineNumber":9,"offsetAndLengths":[[10,4]]},{"preview":" fmt.Println(\"end of main\")","lineNumber":10,"offsetAndLengths":[[24,4]]},{"preview":"func main() {","lineNumber":25,"offsetAndLengths":[[0,9]]},{"preview":"func main() {","lineNumber":51,"offsetAndLengths":[[0,9]]},{"preview":" fmt.Println(\"Welcome to main function\")","lineNumber":52,"offsetAndLengths":[[28,4],[33,4]]},{"preview":"fmt.Println(\"end of main function\")","lineNumber":126,"offsetAndLengths":[[20,4],[25,4]]},{"preview":"package main","lineNumber":78,"offsetAndLengths":[[8,4]]},{"preview":"package main","lineNumber":16,"offsetAndLengths":[[8,4]]},{"preview":"package main","lineNumber":0,"offsetAndLengths":[[8,4]]},{"preview":"package main","lineNumber":36,"offsetAndLengths":[[8,4]]},{"preview":"func dummy_func(ch chan int) {","lineNumber":20,"offsetAndLengths":[[0,4],[11,4]]},{"preview":"func dummy_func(ch chan int) {","lineNumber":43,"offsetAndLengths":[[0,4],[11,4]]},{"preview":"\tgo dummy_func(ch)","lineNumber":64,"offsetAndLengths":[[10,4]]},{"preview":"\tgo dummy_func(ch)","lineNumber":29,"offsetAndLengths":[[10,4]]},{"preview":"func routine1(ch chan int, min int, max int) {","lineNumber":86,"offsetAndLengths":[[0,4]]},{"preview":"func routine2(ch chan int, min int, max int) {","lineNumber":96,"offsetAndLengths":[[0,4]]}]},{"__typename":"FileMatch","repository":{"name":"github.com/GoesToEleven/GolangTraining"},"file":{"path":"14_functions/10_func-expression/01_before-func-expression/main.go","url":"/r/github.com/GoesToEleven/GolangTraining/-/blob/14_functions/10_func-expression/01_before-func-expression/main.go","content":"package main\n\nimport \"fmt\"\n\nfunc greeting() {\n\tfmt.Println(\"Hello world!\")\n}\n\nfunc main() {\n\tgreeting()\n}\n"},"lineMatches":[{"preview":"func main() {","lineNumber":8,"offsetAndLengths":[[0,9]]},{"preview":"package main","lineNumber":0,"offsetAndLengths":[[8,4]]},{"preview":"func greeting() {","lineNumber":4,"offsetAndLengths":[[0,4]]}]},{"__typename":"FileMatch","repository":{"name":"github.com/GoesToEleven/GolangTraining"},"file":{"path":"14_functions/10_func-expression/02_func-expression/main.go","url":"/r/github.com/GoesToEleven/GolangTraining/-/blob/14_functions/10_func-expression/02_func-expression/main.go","content":"package main\n\nimport \"fmt\"\n\nfunc main() {\n\n\tgreeting := func() {\n\t\tfmt.Println(\"Hello world!\")\n\t}\n\n\tgreeting()\n}\n"},"lineMatches":[{"preview":"func main() {","lineNumber":4,"offsetAndLengths":[[0,9]]},{"preview":"\tgreeting := func() {","lineNumber":6,"offsetAndLengths":[[13,4]]}]}]}}}}'

84− headers:

85− Content-Type:

86− - application/json

87− status: 200 OK

88− code: 200

89− duration: 1.268477625s

90−- id: 2

91− request:

92− proto: HTTP/1.1

93− proto_major: 1

94− proto_minor: 1

95− content_length: 44343

96− host: ""

97− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/sourcegraph_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use sourcegraph to search for 'func main' in Go repositories\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_0fb31da44aad4c97a3ccff8f\",\"function\":{\"arguments\":\"{\\\"count\\\": 10, \\\"query\\\": \\\"func main lang:Go\\\"}\",\"name\":\"sourcegraph\"},\"type\":\"function\"}],\"role\":\"assistant\"},{\"content\":\"# Sourcegraph Search Results\\n\\nFound 30 matches across 30 results\\n(Result limit reached, try a more specific query)\\n\\n## Result 1: github.com/jmbharathram/executeoncommand/go/channel/main.go\\n\\nURL: /r/github.com/jmbharathram/executeoncommand/-/blob/go/channel/main.go\\n\\n```\\n96| \\n97| func routine2(ch chan int, min int, max int) {\\n98| \\n99| rand.Seed(time.Now().UnixNano())\\n100| t := rand.Intn(max - min) + min\\n101| fmt.Println(\\\"routine 2 is waiting: \\\", t, \\\"secs\\\")\\n102| time.Sleep(time.Duration(t)*time.Second)\\n103| \\n104| ch <- t\\n105| }\\n106| func main(){\\n107| func main(){\\n108| \\n109| min := 1\\n110| max := 10\\n111| \\n112| ch1:= make(chan int)\\n113| ch2:= make(chan int)\\n114| \\n115| go routine1(ch1, min, max)\\n116| go routine2(ch2, min, max)\\n```\\n\\n```\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| \\n5| func dummy_func() {\\n6| func dummy_func() {\\n7| \\tfmt.Println(\\\"dummy_func\\\")\\n8| }\\n9| func main() {\\n10| \\tgo dummy_func()\\n11| fmt.Println(\\\"end of main\\\")\\n12| }\\n13| \\n14| \\n15| // Example 1\\n```\\n\\n```\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| \\n5| // Example 0\\n6| \\tfmt.Println(\\\"dummy_func\\\")\\n7| \\tfmt.Println(\\\"dummy_func\\\")\\n8| }\\n9| func main() {\\n10| \\tgo dummy_func()\\n11| fmt.Println(\\\"end of main\\\")\\n12| }\\n13| \\n14| \\n15| // Example 1\\n16| \\n```\\n\\n```\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| \\n5| // Example 0\\n6| func dummy_func() {\\n7| \\tfmt.Println(\\\"dummy_func\\\")\\n8| func main() {\\n9| func main() {\\n10| \\tgo dummy_func()\\n11| fmt.Println(\\\"end of main\\\")\\n12| }\\n13| \\n14| \\n15| // Example 1\\n16| \\n17| package main\\n18| \\n```\\n\\n```\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| \\n5| // Example 0\\n6| func dummy_func() {\\n7| \\tfmt.Println(\\\"dummy_func\\\")\\n8| }\\n9| \\tgo dummy_func()\\n10| \\tgo dummy_func()\\n11| fmt.Println(\\\"end of main\\\")\\n12| }\\n13| \\n14| \\n15| // Example 1\\n16| \\n17| package main\\n18| \\n19| import \\\"fmt\\\"\\n```\\n\\n```\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| \\n5| // Example 0\\n6| func dummy_func() {\\n7| \\tfmt.Println(\\\"dummy_func\\\")\\n8| }\\n9| func main() {\\n10| fmt.Println(\\\"end of main\\\")\\n11| fmt.Println(\\\"end of main\\\")\\n12| }\\n13| \\n14| \\n15| // Example 1\\n16| \\n17| package main\\n18| \\n19| import \\\"fmt\\\"\\n20| \\n```\\n\\n```\\n15| // Example 1\\n16| \\n17| package main\\n18| \\n19| import \\\"fmt\\\"\\n20| \\n21| func dummy_func(ch chan int) {\\n22| // receive a value from channel\\n23| val := <-ch\\n24| \\tfmt.Println(val)\\n25| func main() {\\n26| func main() {\\n27| // create a channel\\n28| \\tch := make(chan int)\\n29| // trigger a parallel go routine\\n30| \\tgo dummy_func(ch)\\n31| // send a value into the channel\\n32| \\tch <- 10\\n33| }\\n34| \\n35| // Example 2\\n```\\n\\n```\\n41| //\\\"flag\\\"\\n42| )\\n43| \\n44| func dummy_func(ch chan int) {\\n45| // receive a value from channel\\n46| val := <-ch\\n47| \\tfmt.Println(val)\\n48| \\n49| //send a value again\\n50| ch <- val*2\\n51| func main() {\\n52| func main() {\\n53| fmt.Println(\\\"Welcome to main function\\\")\\n54| \\n55| //flag statements for user input\\n56| var input_value int\\n57| input_value = 10\\n58| //flag.IntVar(&input_value, \\\"v\\\", 10, \\\"Input value to send/receive via channel.\\\" )\\n59| //flag.Parse()\\n60| \\n61| // create a channel\\n```\\n\\n```\\n42| )\\n43| \\n44| func dummy_func(ch chan int) {\\n45| // receive a value from channel\\n46| val := <-ch\\n47| \\tfmt.Println(val)\\n48| \\n49| //send a value again\\n50| ch <- val*2\\n51| }\\n52| fmt.Println(\\\"Welcome to main function\\\")\\n53| fmt.Println(\\\"Welcome to main function\\\")\\n54| \\n55| //flag statements for user input\\n56| var input_value int\\n57| input_value = 10\\n58| //flag.IntVar(&input_value, \\\"v\\\", 10, \\\"Input value to send/receive via channel.\\\" )\\n59| //flag.Parse()\\n60| \\n61| // create a channel\\n62| \\tch := make(chan int)\\n```\\n\\n```\\n116| go routine2(ch2, min, max)\\n117| \\n118| select{\\n119| \\n120| case <- ch1:\\n121| fmt.Println(\\\"routine 1 finished\\\")\\n122| \\n123| case <- ch2:\\n124| fmt.Println(\\\"routine 2 finished\\\")\\n125| }\\n126| fmt.Println(\\\"end of main function\\\")\\n127| fmt.Println(\\\"end of main function\\\")\\n128| \\n129| }\\n130| \\n```\\n\\n```\\n68| \\tch <- input_value\\n69| \\n70| // receive a value from channel\\n71| response := <- ch\\n72| fmt.Println(response)\\n73| \\n74| }\\n75| \\n76| // example 3\\n77| \\n78| package main\\n79| package main\\n80| \\n81| import(\\n82| \\\"fmt\\\"\\n83| \\\"time\\\"\\n84| \\\"math/rand\\\"\\n85| )\\n86| \\n87| func routine1(ch chan int, min int, max int) {\\n88| \\n```\\n\\n```\\n6| func dummy_func() {\\n7| \\tfmt.Println(\\\"dummy_func\\\")\\n8| }\\n9| func main() {\\n10| \\tgo dummy_func()\\n11| fmt.Println(\\\"end of main\\\")\\n12| }\\n13| \\n14| \\n15| // Example 1\\n16| package main\\n17| package main\\n18| \\n19| import \\\"fmt\\\"\\n20| \\n21| func dummy_func(ch chan int) {\\n22| // receive a value from channel\\n23| val := <-ch\\n24| \\tfmt.Println(val)\\n25| }\\n26| func main() {\\n```\\n\\n```\\n0| package main\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| \\n5| // Example 0\\n6| func dummy_func() {\\n7| \\tfmt.Println(\\\"dummy_func\\\")\\n8| }\\n9| func main() {\\n10| \\tgo dummy_func()\\n```\\n\\n```\\n26| func main() {\\n27| // create a channel\\n28| \\tch := make(chan int)\\n29| // trigger a parallel go routine\\n30| \\tgo dummy_func(ch)\\n31| // send a value into the channel\\n32| \\tch <- 10\\n33| }\\n34| \\n35| // Example 2\\n36| package main\\n37| package main\\n38| \\n39| import (\\n40| \\\"fmt\\\"\\n41| //\\\"flag\\\"\\n42| )\\n43| \\n44| func dummy_func(ch chan int) {\\n45| // receive a value from channel\\n46| val := <-ch\\n```\\n\\n```\\n10| \\tgo dummy_func()\\n11| fmt.Println(\\\"end of main\\\")\\n12| }\\n13| \\n14| \\n15| // Example 1\\n16| \\n17| package main\\n18| \\n19| import \\\"fmt\\\"\\n20| func dummy_func(ch chan int) {\\n21| func dummy_func(ch chan int) {\\n22| // receive a value from channel\\n23| val := <-ch\\n24| \\tfmt.Println(val)\\n25| }\\n26| func main() {\\n27| // create a channel\\n28| \\tch := make(chan int)\\n29| // trigger a parallel go routine\\n30| \\tgo dummy_func(ch)\\n```\\n\\n```\\n33| }\\n34| \\n35| // Example 2\\n36| \\n37| package main\\n38| \\n39| import (\\n40| \\\"fmt\\\"\\n41| //\\\"flag\\\"\\n42| )\\n43| func dummy_func(ch chan int) {\\n44| func dummy_func(ch chan int) {\\n45| // receive a value from channel\\n46| val := <-ch\\n47| \\tfmt.Println(val)\\n48| \\n49| //send a value again\\n50| ch <- val*2\\n51| }\\n52| func main() {\\n53| fmt.Println(\\\"Welcome to main function\\\")\\n```\\n\\n```\\n54| \\n55| //flag statements for user input\\n56| var input_value int\\n57| input_value = 10\\n58| //flag.IntVar(&input_value, \\\"v\\\", 10, \\\"Input value to send/receive via channel.\\\" )\\n59| //flag.Parse()\\n60| \\n61| // create a channel\\n62| \\tch := make(chan int)\\n63| \\n64| \\tgo dummy_func(ch)\\n65| \\tgo dummy_func(ch)\\n66| \\n67| // send a value into the channel\\n68| \\tch <- input_value\\n69| \\n70| // receive a value from channel\\n71| response := <- ch\\n72| fmt.Println(response)\\n73| \\n74| }\\n```\\n\\n```\\n19| import \\\"fmt\\\"\\n20| \\n21| func dummy_func(ch chan int) {\\n22| // receive a value from channel\\n23| val := <-ch\\n24| \\tfmt.Println(val)\\n25| }\\n26| func main() {\\n27| // create a channel\\n28| \\tch := make(chan int)\\n29| \\tgo dummy_func(ch)\\n30| \\tgo dummy_func(ch)\\n31| // send a value into the channel\\n32| \\tch <- 10\\n33| }\\n34| \\n35| // Example 2\\n36| \\n37| package main\\n38| \\n39| import (\\n```\\n\\n```\\n76| // example 3\\n77| \\n78| \\n79| package main\\n80| \\n81| import(\\n82| \\\"fmt\\\"\\n83| \\\"time\\\"\\n84| \\\"math/rand\\\"\\n85| )\\n86| func routine1(ch chan int, min int, max int) {\\n87| func routine1(ch chan int, min int, max int) {\\n88| \\n89| rand.Seed(time.Now().UnixNano())\\n90| t := rand.Intn(max - min) + min\\n91| fmt.Println(\\\"routine 1 is waiting: \\\", t, \\\"secs\\\")\\n92| time.Sleep(time.Duration(t)*time.Second)\\n93| \\n94| ch <- t\\n95| }\\n96| \\n```\\n\\n```\\n86| \\n87| func routine1(ch chan int, min int, max int) {\\n88| \\n89| rand.Seed(time.Now().UnixNano())\\n90| t := rand.Intn(max - min) + min\\n91| fmt.Println(\\\"routine 1 is waiting: \\\", t, \\\"secs\\\")\\n92| time.Sleep(time.Duration(t)*time.Second)\\n93| \\n94| ch <- t\\n95| }\\n96| func routine2(ch chan int, min int, max int) {\\n97| func routine2(ch chan int, min int, max int) {\\n98| \\n99| rand.Seed(time.Now().UnixNano())\\n100| t := rand.Intn(max - min) + min\\n101| fmt.Println(\\\"routine 2 is waiting: \\\", t, \\\"secs\\\")\\n102| time.Sleep(time.Duration(t)*time.Second)\\n103| \\n104| ch <- t\\n105| }\\n106| \\n```\\n\\n## Result 2: github.com/GoesToEleven/GolangTraining/14_functions/10_func-expression/01_before-func-expression/main.go\\n\\nURL: /r/github.com/GoesToEleven/GolangTraining/-/blob/14_functions/10_func-expression/01_before-func-expression/main.go\\n\\n```\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| \\n5| func greeting() {\\n6| \\tfmt.Println(\\\"Hello world!\\\")\\n7| }\\n8| func main() {\\n9| func main() {\\n10| \\tgreeting()\\n11| }\\n12| \\n```\\n\\n```\\n0| package main\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| \\n5| func greeting() {\\n6| \\tfmt.Println(\\\"Hello world!\\\")\\n7| }\\n8| \\n9| func main() {\\n10| \\tgreeting()\\n```\\n\\n```\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| func greeting() {\\n5| func greeting() {\\n6| \\tfmt.Println(\\\"Hello world!\\\")\\n7| }\\n8| \\n9| func main() {\\n10| \\tgreeting()\\n11| }\\n12| \\n```\\n\\n## Result 3: github.com/GoesToEleven/GolangTraining/14_functions/10_func-expression/02_func-expression/main.go\\n\\nURL: /r/github.com/GoesToEleven/GolangTraining/-/blob/14_functions/10_func-expression/02_func-expression/main.go\\n\\n```\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| func main() {\\n5| func main() {\\n6| \\n7| \\tgreeting := func() {\\n8| \\t\\tfmt.Println(\\\"Hello world!\\\")\\n9| \\t}\\n10| \\n11| \\tgreeting()\\n12| }\\n13| \\n```\\n\\n```\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| \\n5| func main() {\\n6| \\tgreeting := func() {\\n7| \\tgreeting := func() {\\n8| \\t\\tfmt.Println(\\\"Hello world!\\\")\\n9| \\t}\\n10| \\n11| \\tgreeting()\\n12| }\\n13| \\n```\\n\\n\",\"tool_call_id\":\"call_0fb31da44aad4c97a3ccff8f\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

98− headers:

99− Accept:

100− - application/json

101− Content-Type:

102− - application/json

103− User-Agent:

104− - Charm-Crush/devel (https://charm.land/crush)

105− url: https://hyper.charm.land/v1/chat/completions

106− method: POST

107− response:

108− proto: HTTP/2.0

109− proto_major: 2

110− proto_minor: 0

111− content_length: -1

112− uncompressed: true

113− body: |+

114− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

115

116− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Found"},"finish_reason":null}]}

56+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" same"},"finish_reason":null}]}

117

118− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" 30"},"finish_reason":null}]}

58+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" language"},"finish_reason":null}]}

119

120− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" matches"},"finish_reason":null}]}

60+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

121

122− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" across 30"},"finish_reason":null}]}

62+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"English"},"finish_reason":null}]}

123

124− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" repositories"},"finish_reason":null}]}

64+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")."},"finish_reason":null}]}

125

126− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":". Here"},"finish_reason":null}]}

66+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Summ"},"finish_reason":null}]}

127

128− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" are some"},"finish_reason":null}]}

68+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ar"},"finish_reason":null}]}

129

130− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" notable"},"finish_reason":null}]}

70+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ize"},"finish_reason":null}]}

131

132− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" results:\n\n|"},"finish_reason":null}]}

72+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

133

134− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" # | Repository"},"finish_reason":null}]}

74+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" message"},"finish_reason":null}]}

135

136− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" | File"},"finish_reason":null}]}

76+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

137

138− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" | Line"},"finish_reason":null}]}

78+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

139

140− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" |\n|---"},"finish_reason":null}]}

80+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"use"},"finish_reason":null}]}

141

142− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"|-----------"},"finish_reason":null}]}

82+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" source"},"finish_reason":null}]}

143

144− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"|------"},"finish_reason":null}]}

84+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"graph"},"finish_reason":null}]}

145

146− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"|------|\n"},"finish_reason":null}]}

86+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

147

148− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"| "},"finish_reason":null}]}

88+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" search"},"finish_reason":null}]}

149

150− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"1 | `j"},"finish_reason":null}]}

90+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

151

152− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"mb"},"finish_reason":null}]}

92+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" '"},"finish_reason":null}]}

153

154− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"harath"},"finish_reason":null}]}

94+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"func"},"finish_reason":null}]}

155

156− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"ram/execute"},"finish_reason":null}]}

96+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

157

158− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"on"},"finish_reason":null}]}

98+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"'"},"finish_reason":null}]}

159

160− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"command"},"finish_reason":null}]}

100+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

161

101

162− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"` | `go"},"finish_reason":null}]}

102+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Go"},"finish_reason":null}]}

163

103

164− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"/channel"},"finish_reason":null}]}

104+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" repositories"},"finish_reason":null}]}

165

105

166− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"/main.go"},"finish_reason":null}]}

106+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

167

107

168− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"` | L"},"finish_reason":null}]}

108+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Title"},"finish_reason":null}]}

169

109

170− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"9"},"finish_reason":null}]}

110+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" could"},"finish_reason":null}]}

171

111

172− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":", L"},"finish_reason":null}]}

112+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

173

113

174− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"106"},"finish_reason":null}]}

114+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

175

115

176− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" |\n"},"finish_reason":null}]}

116+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Search"},"finish_reason":null}]}

177

117

178− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"| 2"},"finish_reason":null}]}

118+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

179

119

180− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" | `G"},"finish_reason":null}]}

120+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" func"},"finish_reason":null}]}

181

121

182− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"oesTo"},"finish_reason":null}]}

122+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

183

123

184− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Eleven"},"finish_reason":null}]}

124+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

185

125

186− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"/Golang"},"finish_reason":null}]}

126+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Go"},"finish_reason":null}]}

187

127

188− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Training"},"finish_reason":null}]}

128+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" repos"},"finish_reason":null}]}

189

129

190− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"` | `"},"finish_reason":null}]}

130+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" using"},"finish_reason":null}]}

191

131

192− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"14"},"finish_reason":null}]}

132+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Source"},"finish_reason":null}]}

193

133

194− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"_functions"},"finish_reason":null}]}

134+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"graph"},"finish_reason":null}]}

195

135

196− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"/10"},"finish_reason":null}]}

136+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

197

137

198− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"_func"},"finish_reason":null}]}

138+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

199

139

200− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"-expression"},"finish_reason":null}]}

140+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

201

141

202− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"/01"},"finish_reason":null}]}

142+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

203

143

204− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"_before"},"finish_reason":null}]}

144+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" let's"},"finish_reason":null}]}

205

145

206− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"-f"},"finish_reason":null}]}

146+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

207

147

208− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"unc"},"finish_reason":null}]}

148+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

209

149

210− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"-expression/main"},"finish_reason":null}]}

150+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

211

151

212− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".go` | L"},"finish_reason":null}]}

152+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Search"},"finish_reason":null}]}

213

153

214− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"9 |\n|"},"finish_reason":null}]}

154+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

215

155

216− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" 3"},"finish_reason":null}]}

156+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" func"},"finish_reason":null}]}

217

157

218− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" | `G"},"finish_reason":null}]}

158+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

219

159

220− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"oesToEle"},"finish_reason":null}]}

160+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

221

161

222− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"ven/G"},"finish_reason":null}]}

162+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Go"},"finish_reason":null}]}

223

163

224− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"olangTraining"},"finish_reason":null}]}

164+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" repos"},"finish_reason":null}]}

225

165

226− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"` | `"},"finish_reason":null}]}

166+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" using"},"finish_reason":null}]}

227

167

228− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"14_functions/10"},"finish_reason":null}]}

168+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Source"},"finish_reason":null}]}

229

169

230− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"_func-expression/02"},"finish_reason":null}]}

170+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"graph"},"finish_reason":null}]}

231

171

232− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"_func"},"finish_reason":null}]}

172+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

233

173

234− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"-expression/main"},"finish_reason":null}]}

174+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

235

175

236− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".go"},"finish_reason":null}]}

176+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

237

177

238− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"` | L"},"finish_reason":null}]}

178+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" S"},"finish_reason":null}]}

239

179

240− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"5 |\n"},"finish_reason":null}]}

180+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

241

181

242− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"| 4"},"finish_reason":null}]}

182+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

243

183

244− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" | `G"},"finish_reason":null}]}

184+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

245

185

246− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"oesToEle"},"finish_reason":null}]}

186+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

247

187

248− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"ven/Golang"},"finish_reason":null}]}

188+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"2"},"finish_reason":null}]}

249

189

250− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Training` |"},"finish_reason":null}]}

190+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

251

191

252− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `14_functions"},"finish_reason":null}]}

192+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"3"},"finish_reason":null}]}

253

193

254− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"/10"},"finish_reason":null}]}

194+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" r"},"finish_reason":null}]}

255

195

256− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"_func-expression/..."},"finish_reason":null}]}

196+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

257

197

258− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"` (multiple"},"finish_reason":null}]}

198+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" c"},"finish_reason":null}]}

259

199

260− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" examples"},"finish_reason":null}]}

200+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"5"},"finish_reason":null}]}

261

201

262− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":") | —"},"finish_reason":null}]}

202+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" h"},"finish_reason":null}]}

263

203

264− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" |\n\nThe"},"finish_reason":null}]}

204+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"6"},"finish_reason":null}]}

265

205

266− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" search returned"},"finish_reason":null}]}

206+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

267

207

268− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" results"},"finish_reason":null}]}

208+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

269

209

270− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" predominantly"},"finish_reason":null}]}

210+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"7"},"finish_reason":null}]}

271

211

272− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" from Go"},"finish_reason":null}]}

212+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

273

213

274− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" training/tutorial"},"finish_reason":null}]}

214+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" f"},"finish_reason":null}]}

275

215

276− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" repos"},"finish_reason":null}]}

216+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"8"},"finish_reason":null}]}

277

217

278− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"."},"finish_reason":null}]}

218+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" o"},"finish_reason":null}]}

279

219

280− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" To"},"finish_reason":null}]}

220+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"9"},"finish_reason":null}]}

281

221

282− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" narrow results"},"finish_reason":null}]}

222+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" r"},"finish_reason":null}]}

283

223

284− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" to"},"finish_reason":null}]}

224+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"10"},"finish_reason":null}]}

285

225

286− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" production"},"finish_reason":null}]}

226+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

287

227

288− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" code"},"finish_reason":null}]}

228+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

289

229

290− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":", you"},"finish_reason":null}]}

230+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"11"},"finish_reason":null}]}

291

231

292− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" could refine the query"},"finish_reason":null}]}

232+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

293

233

294− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":","},"finish_reason":null}]}

234+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" f"},"finish_reason":null}]}

295

235

296− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" e.g"},"finish_reason":null}]}

236+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"12"},"finish_reason":null}]}

297

237

298− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".:\n\n-"},"finish_reason":null}]}

238+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" u"},"finish_reason":null}]}

299

239

300− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `func"},"finish_reason":null}]}

240+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"13"},"finish_reason":null}]}

301

241

302− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" main"},"finish_reason":null}]}

242+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" n"},"finish_reason":null}]}

303

243

304− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" lang"},"finish_reason":null}]}

244+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"14"},"finish_reason":null}]}

305

245

306− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":":"},"finish_reason":null}]}

246+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" c"},"finish_reason":null}]}

307

247

308− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Go -"},"finish_reason":null}]}

248+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"15"},"finish_reason":null}]}

309

249

310− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"repo:G"},"finish_reason":null}]}

250+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

311

251

312− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"olang"},"finish_reason":null}]}

252+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

313

253

314− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Training -"},"finish_reason":null}]}

254+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"16"},"finish_reason":null}]}

315

255

316− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"repo:execute"},"finish_reason":null}]}

256+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

317

257

318− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"oncommand`\n"},"finish_reason":null}]}

258+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" m"},"finish_reason":null}]}

319

259

320− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"- `"},"finish_reason":null}]}

260+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"17"},"finish_reason":null}]}

321

261

322− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"func main lang"},"finish_reason":null}]}

262+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

323

263

324− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":":Go repo"},"finish_reason":null}]}

264+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"18"},"finish_reason":null}]}

325

265

326− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":":github"},"finish_reason":null}]}

266+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" i"},"finish_reason":null}]}

327

267

328− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".com"},"finish_reason":null}]}

268+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"19"},"finish_reason":null}]}

329

269

330− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"/s"},"finish_reason":null}]}

270+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" n"},"finish_reason":null}]}

331

271

332− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"ome"},"finish_reason":null}]}

272+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"20"},"finish_reason":null}]}

333

273

334− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"-org"},"finish_reason":null}]}

274+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

335

275

336− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"/s"},"finish_reason":null}]}

276+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

337

277

338− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"ome"},"finish_reason":null}]}

278+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"21"},"finish_reason":null}]}

339

279

340− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"-repo"},"finish_reason":null}]}

280+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

341

281

342− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`\n"},"finish_reason":null}]}

282+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" i"},"finish_reason":null}]}

343

283

344− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"-"},"finish_reason":null}]}

284+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"22"},"finish_reason":null}]}

345

285

346− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

286+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" n"},"finish_reason":null}]}

347

287

348− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"func main lang"},"finish_reason":null}]}

288+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"23"},"finish_reason":null}]}

349

289

350− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":":Go"},"finish_reason":null}]}

290+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

351

291

352− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" file:"},"finish_reason":null}]}

292+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

353

293

354− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"cmd/"},"finish_reason":null}]}

294+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"24"},"finish_reason":null}]}

355

295

356− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`"},"finish_reason":null}]}

296+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

357

297

358− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

298+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" G"},"finish_reason":null}]}

359

299

360− data: {"id":"chatcmpl-f05eca0292d8313c0c109fbb","object":"chat.completion.chunk","created":1779218926,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":3214,"completion_tokens":243,"total_tokens":11265,"prompt_tokens_details":{"cached_tokens":7808},"completion_tokens_details":{"reasoning_tokens":242},"cost":{"usd":0.006955,"hypercredits":0.1391},"remaining":{"hypercredits":0}}}

300+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"25"},"finish_reason":null}]}

361

301

362− data: [DONE]

302+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" o"},"finish_reason":null}]}

363

303

364− headers:

304+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"26"},"finish_reason":null}]}

365− Content-Type:

366− - text/event-stream

367− status: 200 OK

368− code: 200

369− duration: 3.491488583s

370−- id: 3

371− request:

372− proto: HTTP/1.1

373− proto_major: 1

374− proto_minor: 1

375− content_length: 819

376− host: ""

377− body: '{"messages":[{"content":"You will generate a short title based on the first message a user begins a conversation with.\n\n<rules>\n- Keep the title in the same language that the user wrote their message in.\n- Ensure it is not more than 50 characters long.\n- The title should be a summary of the user''s message.\n- It should be one line long.\n- Do not use quotes or colons.\n- The entire text you return will be used as the title.\n- Never return anything that is more than one sentence (one line) long.\n</rules>\n\n /no_think","role":"system"},{"content":"Generate a concise title for the following content:\n\nuse sourcegraph to search for ''func main'' in Go repositories\n <think>\n\n</think>","role":"user"}],"model":"gpt-oss-120b","max_completion_tokens":40,"stream_options":{"include_usage":true},"stream":true}'

378− headers:

379− Accept:

380− - application/json

381− Content-Type:

382− - application/json

383− User-Agent:

384− - Charm-Crush/devel (https://charm.land/crush)

385− url: https://hyper.charm.land/v1/chat/completions

386− method: POST

387− response:

388− proto: HTTP/2.0

389− proto_major: 2

390− proto_minor: 0

391− content_length: -1

392− uncompressed: true

393− body: |+

394− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

395

305

396− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"We"},"finish_reason":null}]}

306+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

397

307

398− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" need"},"finish_reason":null}]}

308+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

399

309

400− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

310+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"27"},"finish_reason":null}]}

401

311

402− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" follow"},"finish_reason":null}]}

312+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

403

313

404− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" developer"},"finish_reason":null}]}

314+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" r"},"finish_reason":null}]}

405

315

406− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" instruction"},"finish_reason":null}]}

316+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"28"},"finish_reason":null}]}

407

317

408− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

318+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

409

319

410− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" generate"},"finish_reason":null}]}

320+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"29"},"finish_reason":null}]}

411

321

412− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

322+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" p"},"finish_reason":null}]}

413

323

414− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" short"},"finish_reason":null}]}

324+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"30"},"finish_reason":null}]}

415

325

416− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

326+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" o"},"finish_reason":null}]}

417

327

418− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" based"},"finish_reason":null}]}

328+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"31"},"finish_reason":null}]}

419

329

420− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" on"},"finish_reason":null}]}

330+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" s"},"finish_reason":null}]}

421

331

422− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

332+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"32"},"finish_reason":null}]}

423

333

424− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" first"},"finish_reason":null}]}

334+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

425

335

426− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" message"},"finish_reason":null}]}

336+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

427

337

428− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

338+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"33"},"finish_reason":null}]}

429

339

430− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" begins"},"finish_reason":null}]}

340+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

431

341

432− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

342+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" u"},"finish_reason":null}]}

433

343

434− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Title"},"finish_reason":null}]}

344+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"34"},"finish_reason":null}]}

435

345

436− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

346+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" s"},"finish_reason":null}]}

437

347

438− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" same"},"finish_reason":null}]}

348+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"35"},"finish_reason":null}]}

439

349

440− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" language"},"finish_reason":null}]}

350+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" i"},"finish_reason":null}]}

441

351

442− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

352+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"36"},"finish_reason":null}]}

443

353

444− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"English"},"finish_reason":null}]}

354+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" n"},"finish_reason":null}]}

445

355

446− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")."},"finish_reason":null}]}

356+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"37"},"finish_reason":null}]}

447

357

448− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" No"},"finish_reason":null}]}

358+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" g"},"finish_reason":null}]}

449

359

450− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" more"},"finish_reason":null}]}

360+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"38"},"finish_reason":null}]}

451

361

452− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" than"},"finish_reason":null}]}

362+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

453

363

454− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

364+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"space"},"finish_reason":null}]}

455

365

456− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

366+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"39"},"finish_reason":null}]}

457

367

458− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

368+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

459

369

460− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

370+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" S"},"finish_reason":null}]}

461

371

462− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" one"},"finish_reason":null}]}

372+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"40"},"finish_reason":null}]}

463

373

464− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

374+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" o"},"finish_reason":null}]}

465

375

466− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

376+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"41"},"finish_reason":null}]}

467

377

468− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

378+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" u"},"finish_reason":null}]}

469

379

470− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

380+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"42"},"finish_reason":null}]}

471

381

472− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" or"},"finish_reason":null}]}

382+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" r"},"finish_reason":null}]}

473

383

474− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" col"},"finish_reason":null}]}

384+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"43"},"finish_reason":null}]}

475

385

476− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ons"},"finish_reason":null}]}

386+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" c"},"finish_reason":null}]}

477

387

478− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

388+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"44"},"finish_reason":null}]}

479

389

480− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" summary"},"finish_reason":null}]}

390+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" e"},"finish_reason":null}]}

481

391

482− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

392+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"45"},"finish_reason":null}]}

483

393

484− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" The"},"finish_reason":null}]}

394+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" g"},"finish_reason":null}]}

485

395

486− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user's"},"finish_reason":null}]}

396+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"46"},"finish_reason":null}]}

487

397

488− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" message"},"finish_reason":null}]}

398+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" r"},"finish_reason":null}]}

489

399

490− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

400+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"47"},"finish_reason":null}]}

491

401

492− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

402+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

493

403

494− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Generate"},"finish_reason":null}]}

404+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"48"},"finish_reason":null}]}

495

405

496− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

406+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" p"},"finish_reason":null}]}

497

407

498− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

408+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"49"},"finish_reason":null}]}

499

409

500− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

410+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" h"},"finish_reason":null}]}

501

411

502− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

412+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

503

413

504− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

414+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

505

415

506− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" following"},"finish_reason":null}]}

416+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Exactly"},"finish_reason":null}]}

507

417

508− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

418+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

509

419

510− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

420+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

511

421

512− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" use"},"finish_reason":null}]}

422+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

513

423

514− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" source"},"finish_reason":null}]}

424+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

515

425

516− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"graph"},"finish_reason":null}]}

426+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Good"},"finish_reason":null}]}

517

427

518− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

428+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

519

429

520− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" search"},"finish_reason":null}]}

430+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" No"},"finish_reason":null}]}

521

431

522− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

432+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

523

433

524− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" '"},"finish_reason":null}]}

434+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" or"},"finish_reason":null}]}

525

435

526− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"func"},"finish_reason":null}]}

436+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" col"},"finish_reason":null}]}

527

437

528− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

438+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ons"},"finish_reason":null}]}

529

439

530− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"'"},"finish_reason":null}]}

440+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

531

441

532− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

442+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" One"},"finish_reason":null}]}

533

443

534− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Go"},"finish_reason":null}]}

444+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

535

445

536− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" repositories"},"finish_reason":null}]}

446+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

537

447

538− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

448+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Search"},"finish_reason":null}]}

539

449

540− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" So"},"finish_reason":null}]}

450+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" for"},"finish_reason":null}]}

541

451

542− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

452+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" func"},"finish_reason":null}]}

543

453

544− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" should"},"finish_reason":null}]}

454+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" main"},"finish_reason":null}]}

545

455

546− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" summarize"},"finish_reason":null}]}

456+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" in"},"finish_reason":null}]}

547

457

548− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" that"},"finish_reason":null}]}

458+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" Go"},"finish_reason":null}]}

549

459

550− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

460+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" repos"},"finish_reason":null}]}

551

461

552− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" maybe"},"finish_reason":null}]}

462+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" using"},"finish_reason":null}]}

553

463

554− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

464+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" Source"},"finish_reason":null}]}

555

465

556− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Using"},"finish_reason":null}]}

466+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"graph"},"finish_reason":null}]}

557

467

558− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Source"},"finish_reason":null}]}

468+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

559

469

560− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"graph"},"finish_reason":null}]}

470+ data: {"id":"chatcmpl-f049264783b8d33719650cf9","object":"chat.completion.chunk","created":1780069701,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":217,"completion_tokens":229,"total_tokens":446,"completion_tokens_details":{"reasoning_tokens":210},"cost":{"usd":0.00030685,"hypercredits":0.006137},"remaining":{"hypercredits":0}}}

561

471

562− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

472+ data: [DONE]

563−

564− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Find"},"finish_reason":null}]}

565−

566− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" func"},"finish_reason":null}]}

567−

568− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

569−

570− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

571

473

572− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Go"},"finish_reason":null}]}

474+ headers:

475+ Content-Type:

476+ - text/event-stream

477+ status: 200 OK

478+ code: 200

479+ duration: 408.3265ms

480+- id: 1

481+ request:

482+ proto: HTTP/1.1

483+ proto_major: 1

484+ proto_minor: 1

485+ content_length: 33958

486+ host: ""

487+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/sourcegraph_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use sourcegraph to search for 'func main' in Go repositories\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

488+ headers:

489+ Accept:

490+ - application/json

491+ Content-Type:

492+ - application/json

493+ User-Agent:

494+ - Charm-Crush/devel (https://charm.land/crush)

495+ url: https://hyper.charm.land/v1/chat/completions

496+ method: POST

497+ response:

498+ proto: HTTP/2.0

499+ proto_major: 2

500+ proto_minor: 0

501+ content_length: -1

502+ uncompressed: true

503+ body: |+

504+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

573

505

574− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

506+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

575

507

576− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Check"},"finish_reason":null}]}

508+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

577

509

578− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" length"},"finish_reason":null}]}

510+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" wants"},"finish_reason":null}]}

579

511

580− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

512+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" me"},"finish_reason":null}]}

581

513

582− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

514+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

583

515

584− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

516+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" use"},"finish_reason":null}]}

585

517

586− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

518+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

587

519

588− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

520+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" Source"},"finish_reason":null}]}

589

521

590− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Using"},"finish_reason":null}]}

522+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"graph"},"finish_reason":null}]}

591

523

592− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Source"},"finish_reason":null}]}

524+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" tool"},"finish_reason":null}]}

593

525

594− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"graph"},"finish_reason":null}]}

526+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

595

527

596− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

528+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" search"},"finish_reason":null}]}

597

529

598− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Find"},"finish_reason":null}]}

530+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

599

531

600− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" func"},"finish_reason":null}]}

532+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" '"},"finish_reason":null}]}

601

533

602− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

534+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"func"},"finish_reason":null}]}

603

535

604− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

536+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

605

537

606− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Go"},"finish_reason":null}]}

538+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"'"},"finish_reason":null}]}

607

539

608− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

540+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

609

541

610− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ="},"finish_reason":null}]}

542+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" Go"},"finish_reason":null}]}

611

543

612− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" let's"},"finish_reason":null}]}

544+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" repositories"},"finish_reason":null}]}

613

545

614− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

546+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

615

547

616− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

548+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_3lLD5JLl0NhdIrtC52F03cgZ","type":"function","index":0,"function":{"name":"sourcegraph","arguments":""}}]},"finish_reason":null}]}

617

549

618− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Using"},"finish_reason":null}]}

550+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

619

551

620− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

552+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

621

553

622− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"5"},"finish_reason":null}]}

554+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"query\": "}}]},"finish_reason":null}]}

623

555

624− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

556+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"func"}}]},"finish_reason":null}]}

625

557

626− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" +"},"finish_reason":null}]}

558+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" main"}}]},"finish_reason":null}]}

627

559

628− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

560+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" lang"}}]},"finish_reason":null}]}

629

561

630− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

562+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":":"}}]},"finish_reason":null}]}

631

563

632− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ="},"finish_reason":null}]}

564+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Go"}}]},"finish_reason":null}]}

633

565

634− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"6"},"finish_reason":null}]}

566+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"}"}}]},"finish_reason":null}]}

635

567

636− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":";"},"finish_reason":null}]}

568+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

637

569

638− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Source"},"finish_reason":null}]}

570+ data: {"id":"chatcmpl-025d8eed69615ae46de3076f","object":"chat.completion.chunk","created":1780069702,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":709,"completion_tokens":37,"total_tokens":7850,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":23},"cost":{"usd":0.0011807900000000001,"hypercredits":0.023615800000000003},"remaining":{"hypercredits":0}}}

639

571

640− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"graph"},"finish_reason":null}]}

572+ data: [DONE]

641

573

642− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

574+ headers:

575+ Content-Type:

576+ - text/event-stream

577+ status: 200 OK

578+ code: 200

579+ duration: 1.04303925s

580+- id: 2

581+ request:

582+ proto: HTTP/1.1

583+ proto_major: 1

584+ proto_minor: 1

585+ content_length: 424

586+ host: sourcegraph.com

587+ body: '{"query":"query Search($query: String!) { search(query: $query, version: V2, patternType: keyword ) { results { matchCount, limitHit, resultCount, approximateResultCount, missing { name }, timedout { name }, indexUnavailable, results { __typename, ... on FileMatch { repository { name }, file { path, url, content }, lineMatches { preview, lineNumber, offsetAndLengths } } } } } }","variables":{"query":"func main lang:Go"}}'

588+ headers:

589+ Content-Type:

590+ - application/json

591+ User-Agent:

592+ - crush/1.0

593+ url: https://sourcegraph.com/.api/graphql

594+ method: POST

595+ response:

596+ proto: HTTP/2.0

597+ proto_major: 2

598+ proto_minor: 0

599+ content_length: -1

600+ uncompressed: true

601+ body: '{"data":{"search":{"results":{"matchCount":30,"limitHit":true,"resultCount":30,"approximateResultCount":"30+","missing":[],"timedout":[],"indexUnavailable":false,"results":[{"__typename":"FileMatch","repository":{"name":"github.com/jmbharathram/executeoncommand"},"file":{"path":"go/channel/main.go","url":"/r/github.com/jmbharathram/executeoncommand/-/blob/go/channel/main.go","content":"package main\n\nimport \"fmt\"\n\n// Example 0\nfunc dummy_func() {\n\tfmt.Println(\"dummy_func\")\n}\nfunc main() {\n\tgo dummy_func()\n fmt.Println(\"end of main\")\n}\n\n\n// Example 1\n\npackage main\n\nimport \"fmt\"\n\nfunc dummy_func(ch chan int) {\n // receive a value from channel\n val := \u003c-ch\n\tfmt.Println(val)\n}\nfunc main() {\n // create a channel\n\tch := make(chan int)\n // trigger a parallel go routine\n\tgo dummy_func(ch)\n // send a value into the channel\n\tch \u003c- 10\n}\n\n// Example 2\n\npackage main\n\nimport (\n \"fmt\"\n //\"flag\"\n)\n\nfunc dummy_func(ch chan int) {\n // receive a value from channel\n val := \u003c-ch\n\tfmt.Println(val)\n\n //send a value again\n ch \u003c- val*2\n}\nfunc main() {\n fmt.Println(\"Welcome to main function\")\n \n //flag statements for user input\n var input_value int\n input_value = 10\n //flag.IntVar(\u0026input_value, \"v\", 10, \"Input value to send/receive via channel.\" )\n //flag.Parse()\n\n // create a channel\n\tch := make(chan int)\n\n // trigger a parallel go routine\n\tgo dummy_func(ch)\n \n // send a value into the channel\n\tch \u003c- input_value\n\n // receive a value from channel\n response := \u003c- ch\n fmt.Println(response)\n\n}\n\n// example 3\n\n\npackage main\n\nimport(\n\"fmt\"\n\"time\"\n\"math/rand\"\n)\n\nfunc routine1(ch chan int, min int, max int) {\n\n rand.Seed(time.Now().UnixNano())\n t := rand.Intn(max - min) + min\n fmt.Println(\"routine 1 is waiting: \", t, \"secs\")\n time.Sleep(time.Duration(t)*time.Second)\n\n ch \u003c- t\n}\n\nfunc routine2(ch chan int, min int, max int) {\n\n rand.Seed(time.Now().UnixNano())\n t := rand.Intn(max - min) + min\n fmt.Println(\"routine 2 is waiting: \", t, \"secs\")\n time.Sleep(time.Duration(t)*time.Second)\n\n ch \u003c- t\n}\n\nfunc main(){\n\nmin := 1\nmax := 10\n\nch1:= make(chan int)\nch2:= make(chan int)\n\ngo routine1(ch1, min, max)\ngo routine2(ch2, min, max)\n\nselect{\n\ncase \u003c- ch1:\n fmt.Println(\"routine 1 finished\")\n\ncase \u003c- ch2:\n fmt.Println(\"routine 2 finished\")\n}\n\nfmt.Println(\"end of main function\")\n\n}\n"},"lineMatches":[{"preview":"func main(){","lineNumber":106,"offsetAndLengths":[[0,9]]},{"preview":"func dummy_func() {","lineNumber":5,"offsetAndLengths":[[0,4],[11,4]]},{"preview":"\tfmt.Println(\"dummy_func\")","lineNumber":6,"offsetAndLengths":[[20,4]]},{"preview":"func main() {","lineNumber":8,"offsetAndLengths":[[0,9]]},{"preview":"\tgo dummy_func()","lineNumber":9,"offsetAndLengths":[[10,4]]},{"preview":" fmt.Println(\"end of main\")","lineNumber":10,"offsetAndLengths":[[24,4]]},{"preview":"func main() {","lineNumber":25,"offsetAndLengths":[[0,9]]},{"preview":"func main() {","lineNumber":51,"offsetAndLengths":[[0,9]]},{"preview":" fmt.Println(\"Welcome to main function\")","lineNumber":52,"offsetAndLengths":[[28,4],[33,4]]},{"preview":"fmt.Println(\"end of main function\")","lineNumber":126,"offsetAndLengths":[[20,4],[25,4]]},{"preview":"package main","lineNumber":78,"offsetAndLengths":[[8,4]]},{"preview":"package main","lineNumber":16,"offsetAndLengths":[[8,4]]},{"preview":"package main","lineNumber":0,"offsetAndLengths":[[8,4]]},{"preview":"package main","lineNumber":36,"offsetAndLengths":[[8,4]]},{"preview":"func dummy_func(ch chan int) {","lineNumber":20,"offsetAndLengths":[[0,4],[11,4]]},{"preview":"func dummy_func(ch chan int) {","lineNumber":43,"offsetAndLengths":[[0,4],[11,4]]},{"preview":"\tgo dummy_func(ch)","lineNumber":64,"offsetAndLengths":[[10,4]]},{"preview":"\tgo dummy_func(ch)","lineNumber":29,"offsetAndLengths":[[10,4]]},{"preview":"func routine1(ch chan int, min int, max int) {","lineNumber":86,"offsetAndLengths":[[0,4]]},{"preview":"func routine2(ch chan int, min int, max int) {","lineNumber":96,"offsetAndLengths":[[0,4]]}]},{"__typename":"FileMatch","repository":{"name":"github.com/GoesToEleven/GolangTraining"},"file":{"path":"14_functions/10_func-expression/01_before-func-expression/main.go","url":"/r/github.com/GoesToEleven/GolangTraining/-/blob/14_functions/10_func-expression/01_before-func-expression/main.go","content":"package main\n\nimport \"fmt\"\n\nfunc greeting() {\n\tfmt.Println(\"Hello world!\")\n}\n\nfunc main() {\n\tgreeting()\n}\n"},"lineMatches":[{"preview":"func main() {","lineNumber":8,"offsetAndLengths":[[0,9]]},{"preview":"package main","lineNumber":0,"offsetAndLengths":[[8,4]]},{"preview":"func greeting() {","lineNumber":4,"offsetAndLengths":[[0,4]]}]},{"__typename":"FileMatch","repository":{"name":"github.com/GoesToEleven/GolangTraining"},"file":{"path":"14_functions/10_func-expression/02_func-expression/main.go","url":"/r/github.com/GoesToEleven/GolangTraining/-/blob/14_functions/10_func-expression/02_func-expression/main.go","content":"package main\n\nimport \"fmt\"\n\nfunc main() {\n\n\tgreeting := func() {\n\t\tfmt.Println(\"Hello world!\")\n\t}\n\n\tgreeting()\n}\n"},"lineMatches":[{"preview":"func main() {","lineNumber":4,"offsetAndLengths":[[0,9]]},{"preview":"\tgreeting := func() {","lineNumber":6,"offsetAndLengths":[[13,4]]}]}]}}}}'

602+ headers:

603+ Content-Type:

604+ - application/json

605+ status: 200 OK

606+ code: 200

607+ duration: 1.390620583s

608+- id: 3

609+ request:

610+ proto: HTTP/1.1

611+ proto_major: 1

612+ proto_minor: 1

613+ content_length: 44423

614+ host: ""

615+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/sourcegraph_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use sourcegraph to search for 'func main' in Go repositories\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_3lLD5JLl0NhdIrtC52F03cgZ\",\"function\":{\"arguments\":\"{\\\"query\\\": \\\"func main lang:Go\\\"}\",\"name\":\"sourcegraph\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"The user wants me to use the Sourcegraph tool to search for 'func main' in Go repositories.\"},{\"content\":\"# Sourcegraph Search Results\\n\\nFound 30 matches across 30 results\\n(Result limit reached, try a more specific query)\\n\\n## Result 1: github.com/jmbharathram/executeoncommand/go/channel/main.go\\n\\nURL: /r/github.com/jmbharathram/executeoncommand/-/blob/go/channel/main.go\\n\\n```\\n96| \\n97| func routine2(ch chan int, min int, max int) {\\n98| \\n99| rand.Seed(time.Now().UnixNano())\\n100| t := rand.Intn(max - min) + min\\n101| fmt.Println(\\\"routine 2 is waiting: \\\", t, \\\"secs\\\")\\n102| time.Sleep(time.Duration(t)*time.Second)\\n103| \\n104| ch <- t\\n105| }\\n106| func main(){\\n107| func main(){\\n108| \\n109| min := 1\\n110| max := 10\\n111| \\n112| ch1:= make(chan int)\\n113| ch2:= make(chan int)\\n114| \\n115| go routine1(ch1, min, max)\\n116| go routine2(ch2, min, max)\\n```\\n\\n```\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| \\n5| func dummy_func() {\\n6| func dummy_func() {\\n7| \\tfmt.Println(\\\"dummy_func\\\")\\n8| }\\n9| func main() {\\n10| \\tgo dummy_func()\\n11| fmt.Println(\\\"end of main\\\")\\n12| }\\n13| \\n14| \\n15| // Example 1\\n```\\n\\n```\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| \\n5| // Example 0\\n6| \\tfmt.Println(\\\"dummy_func\\\")\\n7| \\tfmt.Println(\\\"dummy_func\\\")\\n8| }\\n9| func main() {\\n10| \\tgo dummy_func()\\n11| fmt.Println(\\\"end of main\\\")\\n12| }\\n13| \\n14| \\n15| // Example 1\\n16| \\n```\\n\\n```\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| \\n5| // Example 0\\n6| func dummy_func() {\\n7| \\tfmt.Println(\\\"dummy_func\\\")\\n8| func main() {\\n9| func main() {\\n10| \\tgo dummy_func()\\n11| fmt.Println(\\\"end of main\\\")\\n12| }\\n13| \\n14| \\n15| // Example 1\\n16| \\n17| package main\\n18| \\n```\\n\\n```\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| \\n5| // Example 0\\n6| func dummy_func() {\\n7| \\tfmt.Println(\\\"dummy_func\\\")\\n8| }\\n9| \\tgo dummy_func()\\n10| \\tgo dummy_func()\\n11| fmt.Println(\\\"end of main\\\")\\n12| }\\n13| \\n14| \\n15| // Example 1\\n16| \\n17| package main\\n18| \\n19| import \\\"fmt\\\"\\n```\\n\\n```\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| \\n5| // Example 0\\n6| func dummy_func() {\\n7| \\tfmt.Println(\\\"dummy_func\\\")\\n8| }\\n9| func main() {\\n10| fmt.Println(\\\"end of main\\\")\\n11| fmt.Println(\\\"end of main\\\")\\n12| }\\n13| \\n14| \\n15| // Example 1\\n16| \\n17| package main\\n18| \\n19| import \\\"fmt\\\"\\n20| \\n```\\n\\n```\\n15| // Example 1\\n16| \\n17| package main\\n18| \\n19| import \\\"fmt\\\"\\n20| \\n21| func dummy_func(ch chan int) {\\n22| // receive a value from channel\\n23| val := <-ch\\n24| \\tfmt.Println(val)\\n25| func main() {\\n26| func main() {\\n27| // create a channel\\n28| \\tch := make(chan int)\\n29| // trigger a parallel go routine\\n30| \\tgo dummy_func(ch)\\n31| // send a value into the channel\\n32| \\tch <- 10\\n33| }\\n34| \\n35| // Example 2\\n```\\n\\n```\\n41| //\\\"flag\\\"\\n42| )\\n43| \\n44| func dummy_func(ch chan int) {\\n45| // receive a value from channel\\n46| val := <-ch\\n47| \\tfmt.Println(val)\\n48| \\n49| //send a value again\\n50| ch <- val*2\\n51| func main() {\\n52| func main() {\\n53| fmt.Println(\\\"Welcome to main function\\\")\\n54| \\n55| //flag statements for user input\\n56| var input_value int\\n57| input_value = 10\\n58| //flag.IntVar(&input_value, \\\"v\\\", 10, \\\"Input value to send/receive via channel.\\\" )\\n59| //flag.Parse()\\n60| \\n61| // create a channel\\n```\\n\\n```\\n42| )\\n43| \\n44| func dummy_func(ch chan int) {\\n45| // receive a value from channel\\n46| val := <-ch\\n47| \\tfmt.Println(val)\\n48| \\n49| //send a value again\\n50| ch <- val*2\\n51| }\\n52| fmt.Println(\\\"Welcome to main function\\\")\\n53| fmt.Println(\\\"Welcome to main function\\\")\\n54| \\n55| //flag statements for user input\\n56| var input_value int\\n57| input_value = 10\\n58| //flag.IntVar(&input_value, \\\"v\\\", 10, \\\"Input value to send/receive via channel.\\\" )\\n59| //flag.Parse()\\n60| \\n61| // create a channel\\n62| \\tch := make(chan int)\\n```\\n\\n```\\n116| go routine2(ch2, min, max)\\n117| \\n118| select{\\n119| \\n120| case <- ch1:\\n121| fmt.Println(\\\"routine 1 finished\\\")\\n122| \\n123| case <- ch2:\\n124| fmt.Println(\\\"routine 2 finished\\\")\\n125| }\\n126| fmt.Println(\\\"end of main function\\\")\\n127| fmt.Println(\\\"end of main function\\\")\\n128| \\n129| }\\n130| \\n```\\n\\n```\\n68| \\tch <- input_value\\n69| \\n70| // receive a value from channel\\n71| response := <- ch\\n72| fmt.Println(response)\\n73| \\n74| }\\n75| \\n76| // example 3\\n77| \\n78| package main\\n79| package main\\n80| \\n81| import(\\n82| \\\"fmt\\\"\\n83| \\\"time\\\"\\n84| \\\"math/rand\\\"\\n85| )\\n86| \\n87| func routine1(ch chan int, min int, max int) {\\n88| \\n```\\n\\n```\\n6| func dummy_func() {\\n7| \\tfmt.Println(\\\"dummy_func\\\")\\n8| }\\n9| func main() {\\n10| \\tgo dummy_func()\\n11| fmt.Println(\\\"end of main\\\")\\n12| }\\n13| \\n14| \\n15| // Example 1\\n16| package main\\n17| package main\\n18| \\n19| import \\\"fmt\\\"\\n20| \\n21| func dummy_func(ch chan int) {\\n22| // receive a value from channel\\n23| val := <-ch\\n24| \\tfmt.Println(val)\\n25| }\\n26| func main() {\\n```\\n\\n```\\n0| package main\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| \\n5| // Example 0\\n6| func dummy_func() {\\n7| \\tfmt.Println(\\\"dummy_func\\\")\\n8| }\\n9| func main() {\\n10| \\tgo dummy_func()\\n```\\n\\n```\\n26| func main() {\\n27| // create a channel\\n28| \\tch := make(chan int)\\n29| // trigger a parallel go routine\\n30| \\tgo dummy_func(ch)\\n31| // send a value into the channel\\n32| \\tch <- 10\\n33| }\\n34| \\n35| // Example 2\\n36| package main\\n37| package main\\n38| \\n39| import (\\n40| \\\"fmt\\\"\\n41| //\\\"flag\\\"\\n42| )\\n43| \\n44| func dummy_func(ch chan int) {\\n45| // receive a value from channel\\n46| val := <-ch\\n```\\n\\n```\\n10| \\tgo dummy_func()\\n11| fmt.Println(\\\"end of main\\\")\\n12| }\\n13| \\n14| \\n15| // Example 1\\n16| \\n17| package main\\n18| \\n19| import \\\"fmt\\\"\\n20| func dummy_func(ch chan int) {\\n21| func dummy_func(ch chan int) {\\n22| // receive a value from channel\\n23| val := <-ch\\n24| \\tfmt.Println(val)\\n25| }\\n26| func main() {\\n27| // create a channel\\n28| \\tch := make(chan int)\\n29| // trigger a parallel go routine\\n30| \\tgo dummy_func(ch)\\n```\\n\\n```\\n33| }\\n34| \\n35| // Example 2\\n36| \\n37| package main\\n38| \\n39| import (\\n40| \\\"fmt\\\"\\n41| //\\\"flag\\\"\\n42| )\\n43| func dummy_func(ch chan int) {\\n44| func dummy_func(ch chan int) {\\n45| // receive a value from channel\\n46| val := <-ch\\n47| \\tfmt.Println(val)\\n48| \\n49| //send a value again\\n50| ch <- val*2\\n51| }\\n52| func main() {\\n53| fmt.Println(\\\"Welcome to main function\\\")\\n```\\n\\n```\\n54| \\n55| //flag statements for user input\\n56| var input_value int\\n57| input_value = 10\\n58| //flag.IntVar(&input_value, \\\"v\\\", 10, \\\"Input value to send/receive via channel.\\\" )\\n59| //flag.Parse()\\n60| \\n61| // create a channel\\n62| \\tch := make(chan int)\\n63| \\n64| \\tgo dummy_func(ch)\\n65| \\tgo dummy_func(ch)\\n66| \\n67| // send a value into the channel\\n68| \\tch <- input_value\\n69| \\n70| // receive a value from channel\\n71| response := <- ch\\n72| fmt.Println(response)\\n73| \\n74| }\\n```\\n\\n```\\n19| import \\\"fmt\\\"\\n20| \\n21| func dummy_func(ch chan int) {\\n22| // receive a value from channel\\n23| val := <-ch\\n24| \\tfmt.Println(val)\\n25| }\\n26| func main() {\\n27| // create a channel\\n28| \\tch := make(chan int)\\n29| \\tgo dummy_func(ch)\\n30| \\tgo dummy_func(ch)\\n31| // send a value into the channel\\n32| \\tch <- 10\\n33| }\\n34| \\n35| // Example 2\\n36| \\n37| package main\\n38| \\n39| import (\\n```\\n\\n```\\n76| // example 3\\n77| \\n78| \\n79| package main\\n80| \\n81| import(\\n82| \\\"fmt\\\"\\n83| \\\"time\\\"\\n84| \\\"math/rand\\\"\\n85| )\\n86| func routine1(ch chan int, min int, max int) {\\n87| func routine1(ch chan int, min int, max int) {\\n88| \\n89| rand.Seed(time.Now().UnixNano())\\n90| t := rand.Intn(max - min) + min\\n91| fmt.Println(\\\"routine 1 is waiting: \\\", t, \\\"secs\\\")\\n92| time.Sleep(time.Duration(t)*time.Second)\\n93| \\n94| ch <- t\\n95| }\\n96| \\n```\\n\\n```\\n86| \\n87| func routine1(ch chan int, min int, max int) {\\n88| \\n89| rand.Seed(time.Now().UnixNano())\\n90| t := rand.Intn(max - min) + min\\n91| fmt.Println(\\\"routine 1 is waiting: \\\", t, \\\"secs\\\")\\n92| time.Sleep(time.Duration(t)*time.Second)\\n93| \\n94| ch <- t\\n95| }\\n96| func routine2(ch chan int, min int, max int) {\\n97| func routine2(ch chan int, min int, max int) {\\n98| \\n99| rand.Seed(time.Now().UnixNano())\\n100| t := rand.Intn(max - min) + min\\n101| fmt.Println(\\\"routine 2 is waiting: \\\", t, \\\"secs\\\")\\n102| time.Sleep(time.Duration(t)*time.Second)\\n103| \\n104| ch <- t\\n105| }\\n106| \\n```\\n\\n## Result 2: github.com/GoesToEleven/GolangTraining/14_functions/10_func-expression/01_before-func-expression/main.go\\n\\nURL: /r/github.com/GoesToEleven/GolangTraining/-/blob/14_functions/10_func-expression/01_before-func-expression/main.go\\n\\n```\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| \\n5| func greeting() {\\n6| \\tfmt.Println(\\\"Hello world!\\\")\\n7| }\\n8| func main() {\\n9| func main() {\\n10| \\tgreeting()\\n11| }\\n12| \\n```\\n\\n```\\n0| package main\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| \\n5| func greeting() {\\n6| \\tfmt.Println(\\\"Hello world!\\\")\\n7| }\\n8| \\n9| func main() {\\n10| \\tgreeting()\\n```\\n\\n```\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| func greeting() {\\n5| func greeting() {\\n6| \\tfmt.Println(\\\"Hello world!\\\")\\n7| }\\n8| \\n9| func main() {\\n10| \\tgreeting()\\n11| }\\n12| \\n```\\n\\n## Result 3: github.com/GoesToEleven/GolangTraining/14_functions/10_func-expression/02_func-expression/main.go\\n\\nURL: /r/github.com/GoesToEleven/GolangTraining/-/blob/14_functions/10_func-expression/02_func-expression/main.go\\n\\n```\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| func main() {\\n5| func main() {\\n6| \\n7| \\tgreeting := func() {\\n8| \\t\\tfmt.Println(\\\"Hello world!\\\")\\n9| \\t}\\n10| \\n11| \\tgreeting()\\n12| }\\n13| \\n```\\n\\n```\\n1| package main\\n2| \\n3| import \\\"fmt\\\"\\n4| \\n5| func main() {\\n6| \\tgreeting := func() {\\n7| \\tgreeting := func() {\\n8| \\t\\tfmt.Println(\\\"Hello world!\\\")\\n9| \\t}\\n10| \\n11| \\tgreeting()\\n12| }\\n13| \\n```\\n\\n\",\"tool_call_id\":\"call_3lLD5JLl0NhdIrtC52F03cgZ\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

616+ headers:

617+ Accept:

618+ - application/json

619+ Content-Type:

620+ - application/json

621+ User-Agent:

622+ - Charm-Crush/devel (https://charm.land/crush)

623+ url: https://hyper.charm.land/v1/chat/completions

624+ method: POST

625+ response:

626+ proto: HTTP/2.0

627+ proto_major: 2

628+ proto_minor: 0

629+ content_length: -1

630+ uncompressed: true

631+ body: |+

632+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

643

633

644− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"11"},"finish_reason":null}]}

634+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Found"},"finish_reason":null}]}

645

635

646− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

636+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" "},"finish_reason":null}]}

647

637

648− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" =\u003e"},"finish_reason":null}]}

638+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"30"},"finish_reason":null}]}

649

639

650− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" total"},"finish_reason":null}]}

640+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" matches"},"finish_reason":null}]}

651

641

652− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

642+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" across"},"finish_reason":null}]}

653

643

654− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"17"},"finish_reason":null}]}

644+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" "},"finish_reason":null}]}

655

645

656− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":";"},"finish_reason":null}]}

646+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"30"},"finish_reason":null}]}

657

647

658− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

648+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" results"},"finish_reason":null}]}

659

649

660− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ="},"finish_reason":null}]}

650+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"."},"finish_reason":null}]}

661

651

662− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"18"},"finish_reason":null}]}

652+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" Key"},"finish_reason":null}]}

663

653

664− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":";"},"finish_reason":null}]}

654+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" examples"},"finish_reason":null}]}

665

655

666− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

656+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" include"},"finish_reason":null}]}

667

657

668− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

658+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":":\n\n"},"finish_reason":null}]}

669

659

670− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"2"},"finish_reason":null}]}

660+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"-"},"finish_reason":null}]}

671

661

672− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

662+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" **"},"finish_reason":null}]}

673

663

674− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"20"},"finish_reason":null}]}

664+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"github"},"finish_reason":null}]}

675

665

676− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":";"},"finish_reason":null}]}

666+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".com"},"finish_reason":null}]}

677

667

678− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

668+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"/j"},"finish_reason":null}]}

679

669

680− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

670+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"mb"},"finish_reason":null}]}

681

671

682− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"21"},"finish_reason":null}]}

672+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"har"},"finish_reason":null}]}

683

673

684− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":";"},"finish_reason":null}]}

674+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"ath"},"finish_reason":null}]}

685

675

686− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Find"},"finish_reason":null}]}

676+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"ram"},"finish_reason":null}]}

687

677

688− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

678+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"/"},"finish_reason":null}]}

689

679

690− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

680+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"execute"},"finish_reason":null}]}

691

681

692− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

682+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"on"},"finish_reason":null}]}

693

683

694− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"25"},"finish_reason":null}]}

684+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"command"},"finish_reason":null}]}

695

685

696− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":";"},"finish_reason":null}]}

686+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"**"},"finish_reason":null}]}

697

687

698− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

688+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" —"},"finish_reason":null}]}

699

689

700− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

690+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

701

691

702− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"26"},"finish_reason":null}]}

692+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"main"},"finish_reason":null}]}

703

693

704− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":";"},"finish_reason":null}]}

694+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".go"},"finish_reason":null}]}

705

695

706− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" func"},"finish_reason":null}]}

696+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`"},"finish_reason":null}]}

707

697

708− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

698+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" with"},"finish_reason":null}]}

709

699

710− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

700+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" gor"},"finish_reason":null}]}

711

701

712− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

702+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"outines"},"finish_reason":null}]}

713

703

714− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"30"},"finish_reason":null}]}

704+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" and"},"finish_reason":null}]}

715

705

716− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":";"},"finish_reason":null}]}

706+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" channels"},"finish_reason":null}]}

717

707

718− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

708+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"\n"},"finish_reason":null}]}

719

709

720− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

710+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"-"},"finish_reason":null}]}

721

711

722− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"31"},"finish_reason":null}]}

712+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" **"},"finish_reason":null}]}

723

713

724− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":";"},"finish_reason":null}]}

714+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"github"},"finish_reason":null}]}

725

715

726− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

716+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".com"},"finish_reason":null}]}

727

717

728− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

718+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"/G"},"finish_reason":null}]}

729

719

730− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

720+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"oes"},"finish_reason":null}]}

731

721

732− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

722+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"To"},"finish_reason":null}]}

733

723

734− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"35"},"finish_reason":null}]}

724+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Ele"},"finish_reason":null}]}

735

725

736− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":";"},"finish_reason":null}]}

726+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"ven"},"finish_reason":null}]}

737

727

738− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

728+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"/G"},"finish_reason":null}]}

739

729

740− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

730+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"olang"},"finish_reason":null}]}

741

731

742− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"36"},"finish_reason":null}]}

732+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Training"},"finish_reason":null}]}

743

733

744− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":";"},"finish_reason":null}]}

734+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"**"},"finish_reason":null}]}

745

735

746− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

736+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" —"},"finish_reason":null}]}

747

737

748− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

738+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" Multiple"},"finish_reason":null}]}

749

739

750− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"2"},"finish_reason":null}]}

740+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

751

741

752− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

742+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"func"},"finish_reason":null}]}

753

743

754− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"38"},"finish_reason":null}]}

744+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" main"},"finish_reason":null}]}

755

745

756− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":";"},"finish_reason":null}]}

746+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`"},"finish_reason":null}]}

757

747

758− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

748+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" examples"},"finish_reason":null}]}

759

749

760− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

750+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" demonstrating"},"finish_reason":null}]}

761

751

762− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"39"},"finish_reason":null}]}

752+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" function"},"finish_reason":null}]}

763

753

764− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":";"},"finish_reason":null}]}

754+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" expressions"},"finish_reason":null}]}

765

755

766− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Go"},"finish_reason":null}]}

756+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"\n\n"},"finish_reason":null}]}

767

757

768− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

758+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"The"},"finish_reason":null}]}

769

759

770− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"2"},"finish_reason":null}]}

760+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" results"},"finish_reason":null}]}

771

761

772− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

762+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" are"},"finish_reason":null}]}

773

763

774− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"41"},"finish_reason":null}]}

764+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" mostly"},"finish_reason":null}]}

775

765

776− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

766+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" from"},"finish_reason":null}]}

777

767

778− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

768+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" tutorial"},"finish_reason":null}]}

779

769

780− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

770+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"/tr"},"finish_reason":null}]}

781

771

782− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"41"},"finish_reason":null}]}

772+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"aining"},"finish_reason":null}]}

783

773

784− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

774+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" repos"},"finish_reason":null}]}

785

775

786− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

776+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"."},"finish_reason":null}]}

787

777

788− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" under"},"finish_reason":null}]}

778+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" To"},"finish_reason":null}]}

789

779

790− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

780+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" narrow"},"finish_reason":null}]}

791

781

792− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

782+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" results"},"finish_reason":null}]}

793

783

794− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

784+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":","},"finish_reason":null}]}

795

785

796− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" No"},"finish_reason":null}]}

786+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" try"},"finish_reason":null}]}

797

787

798− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

788+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" a"},"finish_reason":null}]}

799

789

800− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

790+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" more"},"finish_reason":null}]}

801

791

802− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

792+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" specific"},"finish_reason":null}]}

803

793

804− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" colon"},"finish_reason":null}]}

794+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" query"},"finish_reason":null}]}

805

795

806− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

796+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" like"},"finish_reason":null}]}

807

797

808− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Good"},"finish_reason":null}]}

798+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

809

799

810− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

800+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"func"},"finish_reason":null}]}

811

801

812− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Return"},"finish_reason":null}]}

802+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" main"},"finish_reason":null}]}

813

803

814− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" that"},"finish_reason":null}]}

804+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"()"},"finish_reason":null}]}

815

805

816− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

806+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" lang"},"finish_reason":null}]}

817

807

818− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Using"},"finish_reason":null}]}

808+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":":"},"finish_reason":null}]}

819

809

820− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" Source"},"finish_reason":null}]}

810+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Go"},"finish_reason":null}]}

821

811

822− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"graph"},"finish_reason":null}]}

812+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" -"},"finish_reason":null}]}

823

813

824− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" to"},"finish_reason":null}]}

814+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"repo"},"finish_reason":null}]}

825

815

826− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" Find"},"finish_reason":null}]}

816+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":":"},"finish_reason":null}]}

827

817

828− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" func"},"finish_reason":null}]}

818+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"G"},"finish_reason":null}]}

829

819

830− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" main"},"finish_reason":null}]}

820+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"olang"},"finish_reason":null}]}

831

821

832− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" in"},"finish_reason":null}]}

822+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Training"},"finish_reason":null}]}

833

823

834− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" Go"},"finish_reason":null}]}

824+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`."},"finish_reason":null}]}

835

825

836− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

826+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

837

827

838− data: {"id":"chatcmpl-c97c24cf224e9ebefb52fc99","object":"chat.completion.chunk","created":1779218916,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":217,"completion_tokens":220,"total_tokens":437,"completion_tokens_details":{"reasoning_tokens":211},"cost":{"usd":0.0001941,"hypercredits":0.003882},"remaining":{"hypercredits":0}}}

828+ data: {"id":"chatcmpl-438e2c6bfceb138401c90578","object":"chat.completion.chunk","created":1780069704,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":3203,"completion_tokens":98,"total_tokens":11109,"prompt_tokens_details":{"cached_tokens":7808},"cost":{"usd":0.00460753,"hypercredits":0.0921506},"remaining":{"hypercredits":0}}}

839

829

840 data: [DONE]

830 data: [DONE]

841

831

@@ -844,4 +834,4 @@ interactions:

844 - text/event-stream

834 - text/event-stream

845 status: 200 OK

835 status: 200 OK

846 code: 200

836 code: 200

847− duration: 967.360125ms

837+ duration: 769.076542ms

internal/agent/testdata/TestCoderAgent/glm-5.1/update_a_file.yaml+244−346

@@ -25,337 +25,217 @@ interactions:

25 content_length: -1

26 uncompressed: true

27 body: |+

28− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

28+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

30− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"User"},"finish_reason":null}]}

30+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

32− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" wants"},"finish_reason":null}]}

32+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

34− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

34+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" wants"},"finish_reason":null}]}

36− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

36+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

38− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

38+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

40− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

40+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

42− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

42+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

44− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

44+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

46− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

46+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

48− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

48+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

50− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"update"},"finish_reason":null}]}

50+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

52− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

52+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"update"},"finish_reason":null}]}

54− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

54+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

56− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".go"},"finish_reason":null}]}

56+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

58− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" file"},"finish_reason":null}]}

58+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".go"},"finish_reason":null}]}

60− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" by"},"finish_reason":null}]}

60+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" file"},"finish_reason":null}]}

62− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" changing"},"finish_reason":null}]}

62+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" by"},"finish_reason":null}]}

64− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

64+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" changing"},"finish_reason":null}]}

66− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" print"},"finish_reason":null}]}

66+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

68− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

68+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" print"},"finish_reason":null}]}

70− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" say"},"finish_reason":null}]}

70+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

72− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" hello"},"finish_reason":null}]}

72+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" say"},"finish_reason":null}]}

74− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" from"},"finish_reason":null}]}

74+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" hello"},"finish_reason":null}]}

76− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" crush"},"finish_reason":null}]}

76+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" from"},"finish_reason":null}]}

78− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

78+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" crush"},"finish_reason":null}]}

80− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Should"},"finish_reason":null}]}

80+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

82− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

82+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Must"},"finish_reason":null}]}

84− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

84+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" be"},"finish_reason":null}]}

86− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" same"},"finish_reason":null}]}

86+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \u003c="},"finish_reason":null}]}

88− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" language"},"finish_reason":null}]}

88+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

90− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

90+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

92− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"English"},"finish_reason":null}]}

92+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

94− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")."},"finish_reason":null}]}

94+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" one"},"finish_reason":null}]}

96− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" No"},"finish_reason":null}]}

96+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

98− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" more"},"finish_reason":null}]}

98+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

100− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" than"},"finish_reason":null}]}

100+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" no"},"finish_reason":null}]}

101

102− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

102+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

103

104− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

104+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" or"},"finish_reason":null}]}

105

106− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

106+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" col"},"finish_reason":null}]}

107

108− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

108+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ons"},"finish_reason":null}]}

109

110− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Something"},"finish_reason":null}]}

110+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

111

112− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" like"},"finish_reason":null}]}

112+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Title"},"finish_reason":null}]}

113

114− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

114+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" summar"},"finish_reason":null}]}

115

116− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Update"},"finish_reason":null}]}

116+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"izing"},"finish_reason":null}]}

117

118− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

118+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

119

120− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".go"},"finish_reason":null}]}

120+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Something"},"finish_reason":null}]}

121

122− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

122+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" like"},"finish_reason":null}]}

123

124− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" print"},"finish_reason":null}]}

124+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

125

126− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" hello"},"finish_reason":null}]}

126+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Update"},"finish_reason":null}]}

127

128− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" from"},"finish_reason":null}]}

128+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

129

130− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" crush"},"finish_reason":null}]}

130+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".go"},"finish_reason":null}]}

131

132− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

132+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

133

134− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

134+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" print"},"finish_reason":null}]}

135

136− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

136+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" hello"},"finish_reason":null}]}

137

138− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

138+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" from"},"finish_reason":null}]}

139

140− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" let's"},"finish_reason":null}]}

140+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" crush"},"finish_reason":null}]}

141

142− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

142+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

143

144− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

144+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

145

146− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

146+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" under"},"finish_reason":null}]}

147

148− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Update"},"finish_reason":null}]}

148+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

149

150− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

150+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

151

152− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".go"},"finish_reason":null}]}

152+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

153

154− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

154+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

155

156− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" print"},"finish_reason":null}]}

156+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

157

158− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" hello"},"finish_reason":null}]}

158+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

159

160− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" from"},"finish_reason":null}]}

160+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Update"},"finish_reason":null}]}

161

162− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" crush"},"finish_reason":null}]}

162+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

163

164− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

164+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".go"},"finish_reason":null}]}

165

166− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ="},"finish_reason":null}]}

166+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

167

168− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

168+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" print"},"finish_reason":null}]}

169

170− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"40"},"finish_reason":null}]}

170+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" hello"},"finish_reason":null}]}

171

172− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"?"},"finish_reason":null}]}

172+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" from"},"finish_reason":null}]}

173

174− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Count"},"finish_reason":null}]}

174+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" crush"},"finish_reason":null}]}

175

176− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

176+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

177

178− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Update"},"finish_reason":null}]}

178+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ="},"finish_reason":null}]}

179

180− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

180+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" let's"},"finish_reason":null}]}

181

182− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"6"},"finish_reason":null}]}

182+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

183

184− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

184+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

185

186− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" +"},"finish_reason":null}]}

186+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

187

188− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

188+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" U"},"finish_reason":null}]}

189

190− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

190+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

191

192− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ="},"finish_reason":null}]}

192+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

193

194− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"7"},"finish_reason":null}]}

194+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")"},"finish_reason":null}]}

195

196− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

196+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" p"},"finish_reason":null}]}

197

198− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

198+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"..."},"finish_reason":null}]}

199

200− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".go"},"finish_reason":null}]}

200+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" total"},"finish_reason":null}]}

201

202− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

202+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" maybe"},"finish_reason":null}]}

203

204− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"7"},"finish_reason":null}]}

204+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

205

206− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

206+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"38"},"finish_reason":null}]}

207

208− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"14"},"finish_reason":null}]}

208+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

209

210− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

210+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" No"},"finish_reason":null}]}

211

212− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

212+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" quotes"},"finish_reason":null}]}

213

214− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

214+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

215

216− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"15"},"finish_reason":null}]}

216+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Good"},"finish_reason":null}]}

217

218− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

218+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

219

220− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

220+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Update"},"finish_reason":null}]}

221

222− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

222+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" main"},"finish_reason":null}]}

223

224− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"2"},"finish_reason":null}]}

224+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":".go"},"finish_reason":null}]}

225

226− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

226+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" to"},"finish_reason":null}]}

227

228− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"17"},"finish_reason":null}]}

228+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" print"},"finish_reason":null}]}

229

230− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

230+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" hello"},"finish_reason":null}]}

231

232− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

232+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" from"},"finish_reason":null}]}

233

234− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

234+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" crush"},"finish_reason":null}]}

235

236− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"18"},"finish_reason":null}]}

236+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

237

238− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

238+ data: {"id":"chatcmpl-bac0545ec822d29b0d5a7124","object":"chat.completion.chunk","created":1780069674,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":218,"completion_tokens":113,"total_tokens":331,"completion_tokens_details":{"reasoning_tokens":96},"cost":{"usd":0.00016673,"hypercredits":0.0033345999999999996},"remaining":{"hypercredits":0}}}

239−

240− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" print"},"finish_reason":null}]}

241−

242− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

243−

244− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"5"},"finish_reason":null}]}

245−

246− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

247−

248− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"23"},"finish_reason":null}]}

249−

250− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

251−

252− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

253−

254− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

255−

256− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"24"},"finish_reason":null}]}

257−

258− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

259−

260− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" hello"},"finish_reason":null}]}

261−

262− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

263−

264− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"5"},"finish_reason":null}]}

265−

266− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

267−

268− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"29"},"finish_reason":null}]}

269−

270− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

271−

272− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

273−

274− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

275−

276− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"30"},"finish_reason":null}]}

277−

278− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

279−

280− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" from"},"finish_reason":null}]}

281−

282− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

283−

284− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"4"},"finish_reason":null}]}

285−

286− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

287−

288− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"34"},"finish_reason":null}]}

289−

290− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

291−

292− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" space"},"finish_reason":null}]}

293−

294− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"="},"finish_reason":null}]}

295−

296− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"35"},"finish_reason":null}]}

297−

298− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

299−

300− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" crush"},"finish_reason":null}]}

301−

302− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"("},"finish_reason":null}]}

303−

304− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"5"},"finish_reason":null}]}

305−

306− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")="},"finish_reason":null}]}

307−

308− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"40"},"finish_reason":null}]}

309−

310− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

311−

312− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

313−

314− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

315−

316− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"40"},"finish_reason":null}]}

317−

318− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

319−

320− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

321−

322− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" under"},"finish_reason":null}]}

323−

324− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

325−

326− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

327−

328− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

329−

330− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Good"},"finish_reason":null}]}

331−

332− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

333−

334− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Return"},"finish_reason":null}]}

335−

336− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" that"},"finish_reason":null}]}

337−

338− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

339−

340− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Update"},"finish_reason":null}]}

341−

342− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" main"},"finish_reason":null}]}

343−

344− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":".go"},"finish_reason":null}]}

345−

346− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" to"},"finish_reason":null}]}

347−

348− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" print"},"finish_reason":null}]}

349−

350− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" hello"},"finish_reason":null}]}

351−

352− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" from"},"finish_reason":null}]}

353−

354− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" crush"},"finish_reason":null}]}

355−

356− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

357−

358− data: {"id":"chatcmpl-c308ac808f9513b8bf050bcd","object":"chat.completion.chunk","created":1779218859,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":156,"completion_tokens":8,"total_tokens":164,"completion_tokens_details":{"reasoning_tokens":157},"cost":{"usd":0.0001,"hypercredits":0.002},"remaining":{"hypercredits":0}}}

359

239

360 data: [DONE]

240 data: [DONE]

361

241

@@ -364,15 +244,15 @@ interactions:

364 - text/event-stream

244 - text/event-stream

365 status: 200 OK

245 status: 200 OK

366 code: 200

246 code: 200

367− duration: 792.609ms

247+ duration: 484.749333ms

368- id: 1

248- id: 1

369 request:

249 request:

370 proto: HTTP/1.1

250 proto: HTTP/1.1

371 proto_major: 1

251 proto_major: 1

372 proto_minor: 1

252 proto_minor: 1

373− content_length: 33983

253+ content_length: 33964

374 host: ""

254 host: ""

375− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/update_a_file\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"update the main.go file by changing the print to say hello from crush\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

255+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/update_a_file\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"update the main.go file by changing the print to say hello from crush\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

376 headers:

256 headers:

377 Accept:

257 Accept:

378 - application/json

258 - application/json

@@ -389,107 +269,85 @@ interactions:

389 content_length: -1

269 content_length: -1

390 uncompressed: true

270 uncompressed: true

391 body: |+

271 body: |+

392− data: {"id":"chatcmpl-7f806586b867b0ff4b53afe4","object":"chat.completion.chunk","created":1779218861,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

272+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

393

273

394− data: {"id":"chatcmpl-7f806586b867b0ff4b53afe4","object":"chat.completion.chunk","created":1779218861,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"Let"},"finish_reason":null}]}

274+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"Let"},"finish_reason":null}]}

395

275

396− data: {"id":"chatcmpl-7f806586b867b0ff4b53afe4","object":"chat.completion.chunk","created":1779218861,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" me"},"finish_reason":null}]}

276+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" me"},"finish_reason":null}]}

397

277

398− data: {"id":"chatcmpl-7f806586b867b0ff4b53afe4","object":"chat.completion.chunk","created":1779218861,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" find"},"finish_reason":null}]}

278+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" first"},"finish_reason":null}]}

399

279

400− data: {"id":"chatcmpl-7f806586b867b0ff4b53afe4","object":"chat.completion.chunk","created":1779218861,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

280+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" read"},"finish_reason":null}]}

401

281

402− data: {"id":"chatcmpl-7f806586b867b0ff4b53afe4","object":"chat.completion.chunk","created":1779218861,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" read the main.go"},"finish_reason":null}]}

282+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

403

283

404− data: {"id":"chatcmpl-7f806586b867b0ff4b53afe4","object":"chat.completion.chunk","created":1779218861,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" file first."},"finish_reason":null}]}

284+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" main"},"finish_reason":null}]}

405

285

406− data: {"id":"chatcmpl-7f806586b867b0ff4b53afe4","object":"chat.completion.chunk","created":1779218861,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_0a5796b6b25f4b9fa43c81a3","type":"function","index":0,"function":{"name":"glob","arguments":""}}]},"finish_reason":null}]}

286+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":".go"},"finish_reason":null}]}

407

287

408− data: {"id":"chatcmpl-7f806586b867b0ff4b53afe4","object":"chat.completion.chunk","created":1779218861,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

288+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" file"},"finish_reason":null}]}

409

289

410− data: {"id":"chatcmpl-7f806586b867b0ff4b53afe4","object":"chat.completion.chunk","created":1779218861,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

290+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

411

291

412− data: {"id":"chatcmpl-7f806586b867b0ff4b53afe4","object":"chat.completion.chunk","created":1779218861,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"pattern\": "}}]},"finish_reason":null}]}

292+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" see"},"finish_reason":null}]}

413

293

414− data: {"id":"chatcmpl-7f806586b867b0ff4b53afe4","object":"chat.completion.chunk","created":1779218861,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"**/"}}]},"finish_reason":null}]}

294+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" its"},"finish_reason":null}]}

415

295

416− data: {"id":"chatcmpl-7f806586b867b0ff4b53afe4","object":"chat.completion.chunk","created":1779218861,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"main.go\""}}]},"finish_reason":null}]}

296+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" current"},"finish_reason":null}]}

417

297

418− data: {"id":"chatcmpl-7f806586b867b0ff4b53afe4","object":"chat.completion.chunk","created":1779218861,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"}"}}]},"finish_reason":null}]}

298+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

419

299

420− data: {"id":"chatcmpl-7f806586b867b0ff4b53afe4","object":"chat.completion.chunk","created":1779218861,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

300+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

421

301

422− data: {"id":"chatcmpl-7f806586b867b0ff4b53afe4","object":"chat.completion.chunk","created":1779218861,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":715,"completion_tokens":25,"total_tokens":7844,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":11},"cost":{"usd":0.0012309000000000003,"hypercredits":0.024618000000000004},"remaining":{"hypercredits":0}}}

302+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_F9YjcYzHwrHN1Ic3572Oz4Xr","type":"function","index":0,"function":{"name":"view","arguments":""}}]},"finish_reason":null}]}

423

303

424− data: [DONE]

304+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

425

305

426− headers:

306+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

427− Content-Type:

307+

428− - text/event-stream

308+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"file_path\": "}}]},"finish_reason":null}]}

429− status: 200 OK

309+

430− code: 200

310+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"/tmp"}}]},"finish_reason":null}]}

431− duration: 2.484223875s

311+

432−- id: 2

312+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/c"}}]},"finish_reason":null}]}

433− request:

434− proto: HTTP/1.1

435− proto_major: 1

436− proto_minor: 1

437− content_length: 34350

438− host: ""

439− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/update_a_file\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"update the main.go file by changing the print to say hello from crush\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_0a5796b6b25f4b9fa43c81a3\",\"function\":{\"arguments\":\"{\\\"pattern\\\": \\\"**/main.go\\\"}\",\"name\":\"glob\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"Let me find and read the main.go file first.\"},{\"content\":\"/tmp/crush-test/TestCoderAgent/glm-5.1/update_a_file/main.go\",\"tool_call_id\":\"call_0a5796b6b25f4b9fa43c81a3\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

440− headers:

441− Accept:

442− - application/json

443− Content-Type:

444− - application/json

445− User-Agent:

446− - Charm-Crush/devel (https://charm.land/crush)

447− url: https://hyper.charm.land/v1/chat/completions

448− method: POST

449− response:

450− proto: HTTP/2.0

451− proto_major: 2

452− proto_minor: 0

453− content_length: -1

454− uncompressed: true

455− body: |+

456− data: {"id":"chatcmpl-e708d5ca050c9d8bb4af5f79","object":"chat.completion.chunk","created":1779218863,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

457

313

458− data: {"id":"chatcmpl-e708d5ca050c9d8bb4af5f79","object":"chat.completion.chunk","created":1779218863,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_0915f2ff406443098f1aaa79","type":"function","index":0,"function":{"name":"view","arguments":""}}]},"finish_reason":null}]}

314+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"rush"}}]},"finish_reason":null}]}

459

315

460− data: {"id":"chatcmpl-e708d5ca050c9d8bb4af5f79","object":"chat.completion.chunk","created":1779218863,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

316+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-test"}}]},"finish_reason":null}]}

461

317

462− data: {"id":"chatcmpl-e708d5ca050c9d8bb4af5f79","object":"chat.completion.chunk","created":1779218863,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

318+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/Test"}}]},"finish_reason":null}]}

463

319

464− data: {"id":"chatcmpl-e708d5ca050c9d8bb4af5f79","object":"chat.completion.chunk","created":1779218863,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"file_path\": \"/tmp"}}]},"finish_reason":null}]}

320+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Coder"}}]},"finish_reason":null}]}

465

321

466− data: {"id":"chatcmpl-e708d5ca050c9d8bb4af5f79","object":"chat.completion.chunk","created":1779218863,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/c"}}]},"finish_reason":null}]}

322+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Agent"}}]},"finish_reason":null}]}

467

323

468− data: {"id":"chatcmpl-e708d5ca050c9d8bb4af5f79","object":"chat.completion.chunk","created":1779218863,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"rush-test"}}]},"finish_reason":null}]}

324+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/g"}}]},"finish_reason":null}]}

469

325

470− data: {"id":"chatcmpl-e708d5ca050c9d8bb4af5f79","object":"chat.completion.chunk","created":1779218863,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/Test"}}]},"finish_reason":null}]}

326+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"lm"}}]},"finish_reason":null}]}

471

327

472− data: {"id":"chatcmpl-e708d5ca050c9d8bb4af5f79","object":"chat.completion.chunk","created":1779218863,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"CoderAgent"}}]},"finish_reason":null}]}

328+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-"}}]},"finish_reason":null}]}

473

329

474− data: {"id":"chatcmpl-e708d5ca050c9d8bb4af5f79","object":"chat.completion.chunk","created":1779218863,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/glm"}}]},"finish_reason":null}]}

330+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"5"}}]},"finish_reason":null}]}

475

331

476− data: {"id":"chatcmpl-e708d5ca050c9d8bb4af5f79","object":"chat.completion.chunk","created":1779218863,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-5."}}]},"finish_reason":null}]}

332+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"."}}]},"finish_reason":null}]}

477

333

478− data: {"id":"chatcmpl-e708d5ca050c9d8bb4af5f79","object":"chat.completion.chunk","created":1779218863,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"1/update"}}]},"finish_reason":null}]}

334+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"1"}}]},"finish_reason":null}]}

479

335

480− data: {"id":"chatcmpl-e708d5ca050c9d8bb4af5f79","object":"chat.completion.chunk","created":1779218863,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_a"}}]},"finish_reason":null}]}

336+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/update"}}]},"finish_reason":null}]}

481

337

482− data: {"id":"chatcmpl-e708d5ca050c9d8bb4af5f79","object":"chat.completion.chunk","created":1779218863,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_file"}}]},"finish_reason":null}]}

338+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_a"}}]},"finish_reason":null}]}

483

339

484− data: {"id":"chatcmpl-e708d5ca050c9d8bb4af5f79","object":"chat.completion.chunk","created":1779218863,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/main.go"}}]},"finish_reason":null}]}

340+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_file"}}]},"finish_reason":null}]}

485

341

486− data: {"id":"chatcmpl-e708d5ca050c9d8bb4af5f79","object":"chat.completion.chunk","created":1779218863,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\""}}]},"finish_reason":null}]}

342+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/main"}}]},"finish_reason":null}]}

487

343

488− data: {"id":"chatcmpl-e708d5ca050c9d8bb4af5f79","object":"chat.completion.chunk","created":1779218863,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"}"}}]},"finish_reason":null}]}

344+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".go"}}]},"finish_reason":null}]}

489

345

490− data: {"id":"chatcmpl-e708d5ca050c9d8bb4af5f79","object":"chat.completion.chunk","created":1779218863,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

346+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"}"}}]},"finish_reason":null}]}

491

347

492− data: {"id":"chatcmpl-e708d5ca050c9d8bb4af5f79","object":"chat.completion.chunk","created":1779218863,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":47,"completion_tokens":29,"total_tokens":7884,"prompt_tokens_details":{"cached_tokens":7808},"completion_tokens_details":{"reasoning_tokens":28},"cost":{"usd":0.00032130000000000006,"hypercredits":0.006426000000000001},"remaining":{"hypercredits":0}}}

348+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

349+

350+ data: {"id":"chatcmpl-4a778ce8db55e98cf59d1936","object":"chat.completion.chunk","created":1780069674,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":710,"completion_tokens":43,"total_tokens":7857,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":16},"cost":{"usd":0.0011779,"hypercredits":0.023558},"remaining":{"hypercredits":0}}}

493

351

494 data: [DONE]

352 data: [DONE]

495

353

@@ -498,15 +356,15 @@ interactions:

498 - text/event-stream

356 - text/event-stream

499 status: 200 OK

357 status: 200 OK

500 code: 200

358 code: 200

501− duration: 1.486871625s

359+ duration: 749.955292ms

502−- id: 3

360+- id: 2

503 request:

361 request:

504 proto: HTTP/1.1

362 proto: HTTP/1.1

505 proto_major: 1

363 proto_major: 1

506 proto_minor: 1

364 proto_minor: 1

507− content_length: 34826

365+ content_length: 34502

508 host: ""

366 host: ""

509− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/update_a_file\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"update the main.go file by changing the print to say hello from crush\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_0a5796b6b25f4b9fa43c81a3\",\"function\":{\"arguments\":\"{\\\"pattern\\\": \\\"**/main.go\\\"}\",\"name\":\"glob\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"Let me find and read the main.go file first.\"},{\"content\":\"/tmp/crush-test/TestCoderAgent/glm-5.1/update_a_file/main.go\",\"tool_call_id\":\"call_0a5796b6b25f4b9fa43c81a3\",\"role\":\"tool\"},{\"tool_calls\":[{\"id\":\"call_0915f2ff406443098f1aaa79\",\"function\":{\"arguments\":\"{\\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/update_a_file/main.go\\\"}\",\"name\":\"view\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"\"},{\"content\":\"<file>\\n 1|package main\\n 2|\\n 3|import \\\"fmt\\\"\\n 4|\\n 5|func main() {\\n 6|\\tfmt.Println(\\\"Hello, World!\\\")\\n 7|}\\n 8|\\n</file>\\n\",\"tool_call_id\":\"call_0915f2ff406443098f1aaa79\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

367+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/update_a_file\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"update the main.go file by changing the print to say hello from crush\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_F9YjcYzHwrHN1Ic3572Oz4Xr\",\"function\":{\"arguments\":\"{\\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/update_a_file/main.go\\\"}\",\"name\":\"view\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"Let me first read the main.go file to see its current content.\"},{\"content\":\"<file>\\n 1|package main\\n 2|\\n 3|import \\\"fmt\\\"\\n 4|\\n 5|func main() {\\n 6|\\tfmt.Println(\\\"Hello, World!\\\")\\n 7|}\\n 8|\\n</file>\\n\",\"tool_call_id\":\"call_F9YjcYzHwrHN1Ic3572Oz4Xr\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

510 headers:

368 headers:

511 Accept:

369 Accept:

512 - application/json

370 - application/json

@@ -523,65 +381,89 @@ interactions:

523 content_length: -1

381 content_length: -1

524 uncompressed: true

382 uncompressed: true

525 body: |+

383 body: |+

526− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

384+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

385+

386+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_QUQ7vKujNwzMMi6LCjGzqGk4","type":"function","index":0,"function":{"name":"edit","arguments":""}}]},"finish_reason":null}]}

387+

388+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

389+

390+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

391+

392+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"file_path\": "}}]},"finish_reason":null}]}

393+

394+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"/tmp"}}]},"finish_reason":null}]}

395+

396+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/c"}}]},"finish_reason":null}]}

397+

398+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"rush"}}]},"finish_reason":null}]}

527

399

528− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_e27d4ce983e0461293eb54a3","type":"function","index":0,"function":{"name":"edit","arguments":""}}]},"finish_reason":null}]}

400+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-test"}}]},"finish_reason":null}]}

529

401

530− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

402+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/Test"}}]},"finish_reason":null}]}

531

403

532− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

404+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Coder"}}]},"finish_reason":null}]}

533

405

534− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"file_path\": "}}]},"finish_reason":null}]}

406+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Agent"}}]},"finish_reason":null}]}

535

407

536− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"/tmp/c"}}]},"finish_reason":null}]}

408+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/g"}}]},"finish_reason":null}]}

537

409

538− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"rush"}}]},"finish_reason":null}]}

410+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"lm"}}]},"finish_reason":null}]}

539

411

540− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-test"}}]},"finish_reason":null}]}

412+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-"}}]},"finish_reason":null}]}

541

413

542− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/Test"}}]},"finish_reason":null}]}

414+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"5"}}]},"finish_reason":null}]}

543

415

544− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"CoderAgent"}}]},"finish_reason":null}]}

416+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"."}}]},"finish_reason":null}]}

545

417

546− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/glm"}}]},"finish_reason":null}]}

418+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"1"}}]},"finish_reason":null}]}

547

419

548− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-"}}]},"finish_reason":null}]}

420+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/update"}}]},"finish_reason":null}]}

549

421

550− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"5.1"}}]},"finish_reason":null}]}

422+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_a"}}]},"finish_reason":null}]}

551

423

552− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/update"}}]},"finish_reason":null}]}

424+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_file"}}]},"finish_reason":null}]}

553

425

554− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_a"}}]},"finish_reason":null}]}

426+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/main"}}]},"finish_reason":null}]}

555

427

556− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_file"}}]},"finish_reason":null}]}

428+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".go"}}]},"finish_reason":null}]}

557

429

558− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/main.go"}}]},"finish_reason":null}]}

430+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\", \"new_string\": "}}]},"finish_reason":null}]}

559

431

560− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\", "}}]},"finish_reason":null}]}

432+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"\\tfmt"}}]},"finish_reason":null}]}

561

433

562− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"new_string\": \"fmt"}}]},"finish_reason":null}]}

434+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".Println"}}]},"finish_reason":null}]}

563

435

564− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".Println(\\\""}}]},"finish_reason":null}]}

436+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"(\\\""}}]},"finish_reason":null}]}

565

437

566− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"hello"}}]},"finish_reason":null}]}

438+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"hello"}}]},"finish_reason":null}]}

567

439

568− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" from crush"}}]},"finish_reason":null}]}

440+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" from"}}]},"finish_reason":null}]}

569

441

570− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\\\")\", "}}]},"finish_reason":null}]}

442+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" crush"}}]},"finish_reason":null}]}

571

443

572− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"old_string\": "}}]},"finish_reason":null}]}

444+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\\\")"}}]},"finish_reason":null}]}

573

445

574− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"fmt.Println(\\\"Hello"}}]},"finish_reason":null}]}

446+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\", \"old_string\": "}}]},"finish_reason":null}]}

575

447

576− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":", World"}}]},"finish_reason":null}]}

448+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"\\tfmt"}}]},"finish_reason":null}]}

577

449

578− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"!\\\")\""}}]},"finish_reason":null}]}

450+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".Println"}}]},"finish_reason":null}]}

579

451

580− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"}"}}]},"finish_reason":null}]}

452+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"(\\\""}}]},"finish_reason":null}]}

581

453

582− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

454+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Hello"}}]},"finish_reason":null}]}

583

455

584− data: {"id":"chatcmpl-2fc7e159d46284a0d47f319b","object":"chat.completion.chunk","created":1779218865,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":136,"completion_tokens":55,"total_tokens":7999,"prompt_tokens_details":{"cached_tokens":7808},"completion_tokens_details":{"reasoning_tokens":54},"cost":{"usd":0.0006836,"hypercredits":0.013672},"remaining":{"hypercredits":0}}}

456+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":","}}]},"finish_reason":null}]}

457+

458+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" World"}}]},"finish_reason":null}]}

459+

460+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"!\\\")"}}]},"finish_reason":null}]}

461+

462+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"}"}}]},"finish_reason":null}]}

463+

464+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

465+

466+ data: {"id":"chatcmpl-bc1c0706c28fbf93cc7f727d","object":"chat.completion.chunk","created":1780069675,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":95,"completion_tokens":55,"total_tokens":7958,"prompt_tokens_details":{"cached_tokens":7808},"cost":{"usd":0.00035545,"hypercredits":0.007109},"remaining":{"hypercredits":0}}}

585

467

586 data: [DONE]

468 data: [DONE]

587

469

@@ -590,15 +472,15 @@ interactions:

590 - text/event-stream

472 - text/event-stream

591 status: 200 OK

473 status: 200 OK

592 code: 200

474 code: 200

593− duration: 1.631948375s

475+ duration: 721.554959ms

594−- id: 4

476+- id: 3

595 request:

477 request:

596 proto: HTTP/1.1

478 proto: HTTP/1.1

597 proto_major: 1

479 proto_major: 1

598 proto_minor: 1

480 proto_minor: 1

599− content_length: 35365

481+ content_length: 35047

600 host: ""

482 host: ""

601− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/update_a_file\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"update the main.go file by changing the print to say hello from crush\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_0a5796b6b25f4b9fa43c81a3\",\"function\":{\"arguments\":\"{\\\"pattern\\\": \\\"**/main.go\\\"}\",\"name\":\"glob\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"Let me find and read the main.go file first.\"},{\"content\":\"/tmp/crush-test/TestCoderAgent/glm-5.1/update_a_file/main.go\",\"tool_call_id\":\"call_0a5796b6b25f4b9fa43c81a3\",\"role\":\"tool\"},{\"tool_calls\":[{\"id\":\"call_0915f2ff406443098f1aaa79\",\"function\":{\"arguments\":\"{\\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/update_a_file/main.go\\\"}\",\"name\":\"view\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"\"},{\"content\":\"<file>\\n 1|package main\\n 2|\\n 3|import \\\"fmt\\\"\\n 4|\\n 5|func main() {\\n 6|\\tfmt.Println(\\\"Hello, World!\\\")\\n 7|}\\n 8|\\n</file>\\n\",\"tool_call_id\":\"call_0915f2ff406443098f1aaa79\",\"role\":\"tool\"},{\"tool_calls\":[{\"id\":\"call_e27d4ce983e0461293eb54a3\",\"function\":{\"arguments\":\"{\\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/update_a_file/main.go\\\", \\\"new_string\\\": \\\"fmt.Println(\\\\\\\"hello from crush\\\\\\\")\\\", \\\"old_string\\\": \\\"fmt.Println(\\\\\\\"Hello, World!\\\\\\\")\\\"}\",\"name\":\"edit\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"\"},{\"content\":\"<result>\\nContent replaced in file: /tmp/crush-test/TestCoderAgent/glm-5.1/update_a_file/main.go\\n</result>\\n\",\"tool_call_id\":\"call_e27d4ce983e0461293eb54a3\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

483+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/update_a_file\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"update the main.go file by changing the print to say hello from crush\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_F9YjcYzHwrHN1Ic3572Oz4Xr\",\"function\":{\"arguments\":\"{\\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/update_a_file/main.go\\\"}\",\"name\":\"view\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"Let me first read the main.go file to see its current content.\"},{\"content\":\"<file>\\n 1|package main\\n 2|\\n 3|import \\\"fmt\\\"\\n 4|\\n 5|func main() {\\n 6|\\tfmt.Println(\\\"Hello, World!\\\")\\n 7|}\\n 8|\\n</file>\\n\",\"tool_call_id\":\"call_F9YjcYzHwrHN1Ic3572Oz4Xr\",\"role\":\"tool\"},{\"tool_calls\":[{\"id\":\"call_QUQ7vKujNwzMMi6LCjGzqGk4\",\"function\":{\"arguments\":\"{\\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/update_a_file/main.go\\\", \\\"new_string\\\": \\\"\\\\tfmt.Println(\\\\\\\"hello from crush\\\\\\\")\\\", \\\"old_string\\\": \\\"\\\\tfmt.Println(\\\\\\\"Hello, World!\\\\\\\")\\\"}\",\"name\":\"edit\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"\"},{\"content\":\"<result>\\nContent replaced in file: /tmp/crush-test/TestCoderAgent/glm-5.1/update_a_file/main.go\\n</result>\\n\",\"tool_call_id\":\"call_QUQ7vKujNwzMMi6LCjGzqGk4\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

602 headers:

484 headers:

603 Accept:

485 Accept:

604 - application/json

486 - application/json

@@ -615,29 +497,45 @@ interactions:

615 content_length: -1

497 content_length: -1

616 uncompressed: true

498 uncompressed: true

617 body: |+

499 body: |+

618− data: {"id":"chatcmpl-b54da292ad5819b34e4fd52b","object":"chat.completion.chunk","created":1779218868,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

500+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

501+

502+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Done"},"finish_reason":null}]}

503+

504+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"."},"finish_reason":null}]}

505+

506+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" Updated"},"finish_reason":null}]}

507+

508+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

509+

510+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"main"},"finish_reason":null}]}

511+

512+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".go"},"finish_reason":null}]}

513+

514+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":":"},"finish_reason":null}]}

515+

516+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"6"},"finish_reason":null}]}

619

517

620− data: {"id":"chatcmpl-b54da292ad5819b34e4fd52b","object":"chat.completion.chunk","created":1779218868,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Done"},"finish_reason":null}]}

518+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`"},"finish_reason":null}]}

621

519

622− data: {"id":"chatcmpl-b54da292ad5819b34e4fd52b","object":"chat.completion.chunk","created":1779218868,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":". Updated"},"finish_reason":null}]}

520+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" to"},"finish_reason":null}]}

623

521

624− data: {"id":"chatcmpl-b54da292ad5819b34e4fd52b","object":"chat.completion.chunk","created":1779218868,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `main"},"finish_reason":null}]}

522+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" print"},"finish_reason":null}]}

625

523

626− data: {"id":"chatcmpl-b54da292ad5819b34e4fd52b","object":"chat.completion.chunk","created":1779218868,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".go"},"finish_reason":null}]}

524+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `\""},"finish_reason":null}]}

627

525

628− data: {"id":"chatcmpl-b54da292ad5819b34e4fd52b","object":"chat.completion.chunk","created":1779218868,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":":6"},"finish_reason":null}]}

526+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"hello"},"finish_reason":null}]}

629

527

630− data: {"id":"chatcmpl-b54da292ad5819b34e4fd52b","object":"chat.completion.chunk","created":1779218868,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"` to"},"finish_reason":null}]}

528+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" from"},"finish_reason":null}]}

631

529

632− data: {"id":"chatcmpl-b54da292ad5819b34e4fd52b","object":"chat.completion.chunk","created":1779218868,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" print `\""},"finish_reason":null}]}

530+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" crush"},"finish_reason":null}]}

633

531

634− data: {"id":"chatcmpl-b54da292ad5819b34e4fd52b","object":"chat.completion.chunk","created":1779218868,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"hello from"},"finish_reason":null}]}

532+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"\""},"finish_reason":null}]}

635

533

636− data: {"id":"chatcmpl-b54da292ad5819b34e4fd52b","object":"chat.completion.chunk","created":1779218868,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" crush\"`."},"finish_reason":null}]}

534+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`."},"finish_reason":null}]}

637

535

638− data: {"id":"chatcmpl-b54da292ad5819b34e4fd52b","object":"chat.completion.chunk","created":1779218868,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

536+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

639

537

640− data: {"id":"chatcmpl-b54da292ad5819b34e4fd52b","object":"chat.completion.chunk","created":1779218868,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":98,"completion_tokens":19,"total_tokens":8053,"prompt_tokens_details":{"cached_tokens":7936},"completion_tokens_details":{"reasoning_tokens":18},"cost":{"usd":0.00030980000000000005,"hypercredits":0.006196000000000001},"remaining":{"hypercredits":0}}}

538+ data: {"id":"chatcmpl-f6de8499eb3c914cefe756a9","object":"chat.completion.chunk","created":1780069678,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":57,"completion_tokens":19,"total_tokens":8012,"prompt_tokens_details":{"cached_tokens":7936},"cost":{"usd":0.00015447,"hypercredits":0.0030894},"remaining":{"hypercredits":0}}}

641

539

642 data: [DONE]

540 data: [DONE]

643

541

@@ -646,4 +544,4 @@ interactions:

646 - text/event-stream

544 - text/event-stream

647 status: 200 OK

545 status: 200 OK

648 code: 200

546 code: 200

649− duration: 2.180037s

547+ duration: 2.1932855s

internal/agent/testdata/TestCoderAgent/glm-5.1/write_tool.yaml+204−144

@@ -25,211 +25,219 @@ interactions:

25 content_length: -1

26 uncompressed: true

27 body: |+

28− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

28+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

30− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

30+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"The"},"finish_reason":null}]}

32− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

32+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" user"},"finish_reason":null}]}

34− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" wants"},"finish_reason":null}]}

34+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" asks"},"finish_reason":null}]}

36− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

36+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

38− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

38+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

40− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

40+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Generate"},"finish_reason":null}]}

42− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

42+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

44− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

44+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" concise"},"finish_reason":null}]}

46− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

46+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

48− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

48+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" for"},"finish_reason":null}]}

50− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

50+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" the"},"finish_reason":null}]}

52− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"use"},"finish_reason":null}]}

52+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" following"},"finish_reason":null}]}

54− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" write"},"finish_reason":null}]}

54+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

56− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

56+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":\n\n"},"finish_reason":null}]}

58− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" create"},"finish_reason":null}]}

58+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"use"},"finish_reason":null}]}

60− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

60+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" write"},"finish_reason":null}]}

62− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" new"},"finish_reason":null}]}

62+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

64− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" file"},"finish_reason":null}]}

64+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" create"},"finish_reason":null}]}

66− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" called"},"finish_reason":null}]}

66+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

68− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" config"},"finish_reason":null}]}

68+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" new"},"finish_reason":null}]}

70− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".json"},"finish_reason":null}]}

70+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" file"},"finish_reason":null}]}

72− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

72+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" called"},"finish_reason":null}]}

74− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

74+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" config"},"finish_reason":null}]}

76− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" '{\""},"finish_reason":null}]}

76+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".json"},"finish_reason":null}]}

78− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"name"},"finish_reason":null}]}

78+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

80− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\":"},"finish_reason":null}]}

80+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" content"},"finish_reason":null}]}

82− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

82+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" '{\""},"finish_reason":null}]}

84− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"test"},"finish_reason":null}]}

84+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"name"},"finish_reason":null}]}

86− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\","},"finish_reason":null}]}

86+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\":"},"finish_reason":null}]}

88− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

88+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

90− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"version"},"finish_reason":null}]}

90+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"test"},"finish_reason":null}]}

92− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\":"},"finish_reason":null}]}

92+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\","},"finish_reason":null}]}

94− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

94+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

96− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

96+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"version"},"finish_reason":null}]}

98− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

98+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\":"},"finish_reason":null}]}

100− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"0"},"finish_reason":null}]}

100+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

101

102− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

102+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"1"},"finish_reason":null}]}

103

104− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"0"},"finish_reason":null}]}

104+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

105

106− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"}"},"finish_reason":null}]}

106+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"0"},"finish_reason":null}]}

107

108− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"'"},"finish_reason":null}]}

108+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

109

110− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"."},"finish_reason":null}]}

110+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"0"},"finish_reason":null}]}

111

112− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Title"},"finish_reason":null}]}

112+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"}"},"finish_reason":null}]}

113

114− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" in"},"finish_reason":null}]}

114+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"'"},"finish_reason":null}]}

115

116− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" same"},"finish_reason":null}]}

116+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\"\n\n"},"finish_reason":null}]}

117

118− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" language"},"finish_reason":null}]}

118+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"We"},"finish_reason":null}]}

119

120− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

120+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" need"},"finish_reason":null}]}

121

122− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"English"},"finish_reason":null}]}

122+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

123

124− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":")."},"finish_reason":null}]}

124+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" short"},"finish_reason":null}]}

125

126− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Under"},"finish_reason":null}]}

126+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" title"},"finish_reason":null}]}

127

128− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

128+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

129

130− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

130+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \u003c="},"finish_reason":null}]}

131

132− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

132+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"50"},"finish_reason":null}]}

133

134− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

134+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" chars"},"finish_reason":null}]}

135

136− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Summ"},"finish_reason":null}]}

136+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":","},"finish_reason":null}]}

137

138− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ar"},"finish_reason":null}]}

138+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" same"},"finish_reason":null}]}

139

140− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"ize"},"finish_reason":null}]}

140+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" language"},"finish_reason":null}]}

141

142− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

142+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" ("},"finish_reason":null}]}

143

144− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Something"},"finish_reason":null}]}

144+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"English"},"finish_reason":null}]}

145

146− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" like"},"finish_reason":null}]}

146+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"),"},"finish_reason":null}]}

147

148− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

148+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" summary"},"finish_reason":null}]}

149

150− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Create"},"finish_reason":null}]}

150+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

151

152− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" config"},"finish_reason":null}]}

152+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Something"},"finish_reason":null}]}

153

154− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".json"},"finish_reason":null}]}

154+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" like"},"finish_reason":null}]}

155

156− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

156+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

157

158− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" name"},"finish_reason":null}]}

158+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Create"},"finish_reason":null}]}

159

160− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

160+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" config"},"finish_reason":null}]}

161

162− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" version"},"finish_reason":null}]}

162+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".json"},"finish_reason":null}]}

163

164− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

164+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

165

166− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Let's"},"finish_reason":null}]}

166+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" name"},"finish_reason":null}]}

167

168− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

168+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

169

170− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

170+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" version"},"finish_reason":null}]}

171

172− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

172+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

173

174− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Create"},"finish_reason":null}]}

174+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Let's"},"finish_reason":null}]}

175

176− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" config"},"finish_reason":null}]}

176+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" count"},"finish_reason":null}]}

177

178− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".json"},"finish_reason":null}]}

178+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

179

180− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

180+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" \""},"finish_reason":null}]}

181

182− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" name"},"finish_reason":null}]}

182+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Create"},"finish_reason":null}]}

183

184− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

184+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" config"},"finish_reason":null}]}

185

186− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" version"},"finish_reason":null}]}

186+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".json"},"finish_reason":null}]}

187

188− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

188+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

189

190− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" length"},"finish_reason":null}]}

190+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" name"},"finish_reason":null}]}

191

192− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":":"},"finish_reason":null}]}

192+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" and"},"finish_reason":null}]}

193

194− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

194+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" version"},"finish_reason":null}]}

195

196− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"38"},"finish_reason":null}]}

196+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"\""},"finish_reason":null}]}

197

198− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

198+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" is"},"finish_reason":null}]}

199

200− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

200+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" "},"finish_reason":null}]}

201

202− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Good"},"finish_reason":null}]}

202+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"38"},"finish_reason":null}]}

203

204− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

204+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" characters"},"finish_reason":null}]}

205

206− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" Return"},"finish_reason":null}]}

206+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

207

208− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" just"},"finish_reason":null}]}

208+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" That's"},"finish_reason":null}]}

209

210− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" that"},"finish_reason":null}]}

210+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" fine"},"finish_reason":null}]}

211

212− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

212+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":".\n\n"},"finish_reason":null}]}

213

214− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

214+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"Return"},"finish_reason":null}]}

215

216− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Create"},"finish_reason":null}]}

216+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" just"},"finish_reason":null}]}

217

218− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" config"},"finish_reason":null}]}

218+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" that"},"finish_reason":null}]}

219

220− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":".json"},"finish_reason":null}]}

220+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":" line"},"finish_reason":null}]}

221

222− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" with"},"finish_reason":null}]}

222+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

223

224− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" name"},"finish_reason":null}]}

224+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":"Create"},"finish_reason":null}]}

225

226− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" and"},"finish_reason":null}]}

226+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" config"},"finish_reason":null}]}

227

228− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" version"},"finish_reason":null}]}

228+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":".json"},"finish_reason":null}]}

229

230− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

230+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" with"},"finish_reason":null}]}

231

232− data: {"id":"chatcmpl-4dac3d240450caf171112238","object":"chat.completion.chunk","created":1779218932,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":232,"completion_tokens":111,"total_tokens":343,"completion_tokens_details":{"reasoning_tokens":94},"cost":{"usd":0.00010520000000000001,"hypercredits":0.002104},"remaining":{"hypercredits":0}}}

232+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" name"},"finish_reason":null}]}

233+

234+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" and"},"finish_reason":null}]}

235+

236+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{"content":" version"},"finish_reason":null}]}

237+

238+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

239+

240+ data: {"id":"chatcmpl-8ffea572a2477bb64ba0a191","object":"chat.completion.chunk","created":1780069705,"model":"gpt-oss-120b","choices":[],"usage":{"prompt_tokens":232,"completion_tokens":115,"total_tokens":347,"completion_tokens_details":{"reasoning_tokens":98},"cost":{"usd":0.00017169,"hypercredits":0.0034338},"remaining":{"hypercredits":0}}}

233

241

234 data: [DONE]

242 data: [DONE]

235

243

@@ -238,15 +246,15 @@ interactions:

238 - text/event-stream

246 - text/event-stream

239 status: 200 OK

247 status: 200 OK

240 code: 200

248 code: 200

241− duration: 432.121292ms

249+ duration: 396.384ms

242- id: 1

250- id: 1

243 request:

251 request:

244 proto: HTTP/1.1

252 proto: HTTP/1.1

245 proto_major: 1

253 proto_major: 1

246 proto_minor: 1

254 proto_minor: 1

247− content_length: 34020

255+ content_length: 34001

248 host: ""

256 host: ""

249− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/write_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use write to create a new file called config.json with content '{\\\"name\\\": \\\"test\\\", \\\"version\\\": \\\"1.0.0\\\"}'\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

257+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/write_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use write to create a new file called config.json with content '{\\\"name\\\": \\\"test\\\", \\\"version\\\": \\\"1.0.0\\\"}'\",\"role\":\"user\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

250 headers:

258 headers:

251 Accept:

259 Accept:

252 - application/json

260 - application/json

@@ -263,63 +271,109 @@ interactions:

263 content_length: -1

271 content_length: -1

264 uncompressed: true

272 uncompressed: true

265 body: |+

273 body: |+

266− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

274+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

275+

276+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"Simple"},"finish_reason":null}]}

277+

278+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" request"},"finish_reason":null}]}

279+

280+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" to"},"finish_reason":null}]}

281+

282+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" create"},"finish_reason":null}]}

283+

284+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" a"},"finish_reason":null}]}

285+

286+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" file"},"finish_reason":null}]}

287+

288+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" with"},"finish_reason":null}]}

289+

290+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" write"},"finish_reason":null}]}

291+

292+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":" tool"},"finish_reason":null}]}

293+

294+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"reasoning_content":"."},"finish_reason":null}]}

295+

296+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_kLqkWWD0dwlybPAAd0yUiCAx","type":"function","index":0,"function":{"name":"write","arguments":""}}]},"finish_reason":null}]}

267

297

268− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"id":"call_59425ba8906d4a64b1c1511d","type":"function","index":0,"function":{"name":"write","arguments":""}}]},"finish_reason":null}]}

298+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

269

299

270− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":""}}]},"finish_reason":null}]}

300+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

271

301

272− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"{"}}]},"finish_reason":null}]}

302+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"content\": "}}]},"finish_reason":null}]}

273

303

274− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"content\": "}}]},"finish_reason":null}]}

304+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"{\\\""}}]},"finish_reason":null}]}

275

305

276− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"{\\\"name"}}]},"finish_reason":null}]}

306+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"name"}}]},"finish_reason":null}]}

277

307

278− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\\\": \\\""}}]},"finish_reason":null}]}

308+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\\\":"}}]},"finish_reason":null}]}

279

309

280− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"test\\\", \\\"version"}}]},"finish_reason":null}]}

310+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" \\\""}}]},"finish_reason":null}]}

281

311

282− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\\\": \\\""}}]},"finish_reason":null}]}

312+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"test"}}]},"finish_reason":null}]}

283

313

284− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"1"}}]},"finish_reason":null}]}

314+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\\\","}}]},"finish_reason":null}]}

285

315

286− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".0"}}]},"finish_reason":null}]}

316+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" \\\""}}]},"finish_reason":null}]}

287

317

288− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".0"}}]},"finish_reason":null}]}

318+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"version"}}]},"finish_reason":null}]}

289

319

290− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\\\"}\""}}]},"finish_reason":null}]}

320+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\\\":"}}]},"finish_reason":null}]}

291

321

292− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":", "}}]},"finish_reason":null}]}

322+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":" \\\""}}]},"finish_reason":null}]}

293

323

294− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"file_path\": "}}]},"finish_reason":null}]}

324+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"1"}}]},"finish_reason":null}]}

295

325

296− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"/tmp/c"}}]},"finish_reason":null}]}

326+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"."}}]},"finish_reason":null}]}

297

327

298− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"rush"}}]},"finish_reason":null}]}

328+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"0"}}]},"finish_reason":null}]}

299

329

300− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-test"}}]},"finish_reason":null}]}

330+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"."}}]},"finish_reason":null}]}

301

331

302− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/Test"}}]},"finish_reason":null}]}

332+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"0"}}]},"finish_reason":null}]}

303

333

304− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Coder"}}]},"finish_reason":null}]}

334+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\\\"}"}}]},"finish_reason":null}]}

305

335

306− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Agent"}}]},"finish_reason":null}]}

336+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\", \"file_path\": "}}]},"finish_reason":null}]}

307

337

308− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/glm"}}]},"finish_reason":null}]}

338+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"/tmp"}}]},"finish_reason":null}]}

309

339

310− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-5"}}]},"finish_reason":null}]}

340+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/c"}}]},"finish_reason":null}]}

311

341

312− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".1/write"}}]},"finish_reason":null}]}

342+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"rush"}}]},"finish_reason":null}]}

313

343

314− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_tool/config"}}]},"finish_reason":null}]}

344+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-test"}}]},"finish_reason":null}]}

315

345

316− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".json\""}}]},"finish_reason":null}]}

346+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/Test"}}]},"finish_reason":null}]}

317

347

318− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"}"}}]},"finish_reason":null}]}

348+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Coder"}}]},"finish_reason":null}]}

319

349

320− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

350+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"Agent"}}]},"finish_reason":null}]}

321

351

322− data: {"id":"chatcmpl-e2dc8bea98612894ac590c36","object":"chat.completion.chunk","created":1779218934,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":729,"completion_tokens":49,"total_tokens":7882,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":48},"cost":{"usd":0.0015203,"hypercredits":0.030406},"remaining":{"hypercredits":0}}}

352+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/g"}}]},"finish_reason":null}]}

353+

354+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"lm"}}]},"finish_reason":null}]}

355+

356+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"-"}}]},"finish_reason":null}]}

357+

358+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"5"}}]},"finish_reason":null}]}

359+

360+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"."}}]},"finish_reason":null}]}

361+

362+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"1"}}]},"finish_reason":null}]}

363+

364+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/write"}}]},"finish_reason":null}]}

365+

366+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"_tool"}}]},"finish_reason":null}]}

367+

368+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"/config"}}]},"finish_reason":null}]}

369+

370+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":".json"}}]},"finish_reason":null}]}

371+

372+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"function":{"arguments":"\"}"}}]},"finish_reason":null}]}

373+

374+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"tool_calls"}]}

375+

376+ data: {"id":"chatcmpl-b2fc9d4641c9729689320104","object":"chat.completion.chunk","created":1780069705,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":724,"completion_tokens":59,"total_tokens":7887,"prompt_tokens_details":{"cached_tokens":7104},"completion_tokens_details":{"reasoning_tokens":12},"cost":{"usd":0.00124664,"hypercredits":0.024932799999999998},"remaining":{"hypercredits":0}}}

323

377

324 data: [DONE]

378 data: [DONE]

325

379

@@ -328,15 +382,15 @@ interactions:

328 - text/event-stream

382 - text/event-stream

329 status: 200 OK

383 status: 200 OK

330 code: 200

384 code: 200

331− duration: 2.341392s

385+ duration: 533.025375ms

332- id: 2

386- id: 2

333 request:

387 request:

334 proto: HTTP/1.1

388 proto: HTTP/1.1

335 proto_major: 1

389 proto_major: 1

336 proto_minor: 1

390 proto_minor: 1

337− content_length: 34502

391+ content_length: 34554

338 host: ""

392 host: ""

339− body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/write_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use write to create a new file called config.json with content '{\\\"name\\\": \\\"test\\\", \\\"version\\\": \\\"1.0.0\\\"}'\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_59425ba8906d4a64b1c1511d\",\"function\":{\"arguments\":\"{\\\"content\\\": \\\"{\\\\\\\"name\\\\\\\": \\\\\\\"test\\\\\\\", \\\\\\\"version\\\\\\\": \\\\\\\"1.0.0\\\\\\\"}\\\", \\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/write_tool/config.json\\\"}\",\"name\":\"write\"},\"type\":\"function\"}],\"role\":\"assistant\"},{\"content\":\"<result>\\nFile successfully written: /tmp/crush-test/TestCoderAgent/glm-5.1/write_tool/config.json\\n</result>\",\"tool_call_id\":\"call_59425ba8906d4a64b1c1511d\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

393+ body: "{\"messages\":[{\"content\":\"You are Crush, a powerful AI Assistant that runs in the CLI.\\n\\n<critical_rules>\\nThese rules override everything else. Follow them strictly:\\n\\n1. **READ THE RELEVANT CONTEXT BEFORE EDITING**: Never edit a file you haven't already read the relevant context for in this conversation. Once read, you don't need to re-read unless it changed. Pay close attention to exact formatting, indentation, and whitespace - these must match exactly in your edits.\\n2. **BE AUTONOMOUS**: Don't ask questions - search, read, think, decide, act. Break complex tasks into steps and complete them all. Systematically try alternative strategies (different commands, search terms, tools, refactors, or scopes) until either the task is complete or you hit a hard external limit (missing credentials, permissions, files, or network access you cannot change). Only stop for actual blocking errors, not perceived difficulty.\\n3. **TEST AFTER CHANGES**: Run tests immediately after each modification.\\n4. **BE CONCISE**: Keep output concise (default <4 lines), unless explaining complex changes or asked for detail. Conciseness applies to output only, not to thoroughness of work.\\n5. **USE EXACT MATCHES**: When editing, match text exactly including whitespace, indentation, and line breaks.\\n6. **NEVER COMMIT**: Unless user explicitly says \\\"commit\\\". When committing, follow the `<git_commits>` format from the bash tool description exactly, including any configured attribution lines.\\n7. **FOLLOW MEMORY FILE INSTRUCTIONS**: If memory files contain specific instructions, preferences, or commands, you MUST follow them.\\n8. **NEVER ADD COMMENTS**: Only add comments if the user asked you to do so. Focus on *why* not *what*. NEVER communicate with the user through code comments.\\n9. **SECURITY FIRST**: Only assist with defensive security tasks. Refuse to create, modify, or improve code that may be used maliciously.\\n10. **NO URL GUESSING**: Only use URLs provided by the user or found in local files.\\n11. **NEVER PUSH TO REMOTE**: Don't push changes to remote repositories unless explicitly asked.\\n12. **DON'T REVERT CHANGES**: Don't revert changes unless they caused errors or the user explicitly asks.\\n13. **TOOL CONSTRAINTS**: Only use documented tools. Never attempt 'apply_patch' or 'apply_diff' - they don't exist. Use 'edit' or 'multiedit' instead.\\n14. **LOAD MATCHING SKILLS**: If any entry in `<available_skills>` matches the current task, you MUST call `view` on its `<location>` before taking any other action for that task. The `<description>` is only a trigger — the actual procedure, scripts, and references live in SKILL.md. Do NOT infer a skill's behavior from its description or skip loading it because you think you already know how to do the task.\\n15. **LIMIT FILE READS**: Avoid reading entire files, as they can be very large. Read only the sections you need using 'offset' and 'limit' parameters.\\n</critical_rules>\\n\\n<communication_style>\\nKeep responses minimal:\\n- ALWAYS think and respond in the same spoken language the prompt was written in.\\n- Under 4 lines of text (tool use doesn't count)\\n- Conciseness is about **text only**: always fully implement the requested feature, tests, and wiring even if that requires many tool calls.\\n- No preamble (\\\"Here's...\\\", \\\"I'll...\\\")\\n- No postamble (\\\"Let me know...\\\", \\\"Hope this helps...\\\")\\n- One-word answers when possible\\n- No emojis ever\\n- No explanations unless user asks\\n- Never send acknowledgement-only responses; after receiving new context or instructions, immediately continue the task or state the concrete next action you will take.\\n- Use rich Markdown formatting (headings, bullet lists, tables, code fences) for any multi-sentence or explanatory answer; only use plain unformatted text if the user explicitly asks.\\n\\nExamples:\\nuser: what is 2+2?\\nassistant: 4\\n\\nuser: list files in src/\\nassistant: [uses ls tool]\\nfoo.c, bar.c, baz.c\\n\\nuser: which file has the foo implementation?\\nassistant: src/foo.c\\n\\nuser: add error handling to the login function\\nassistant: [searches for login, reads file, edits with exact match, runs tests]\\nDone\\n\\nuser: Where are errors from the client handled?\\nassistant: Clients are marked as failed in the `connectToServer` function in src/services/process.go:712.\\n</communication_style>\\n\\n<code_references>\\nWhen referencing specific functions or code locations, use the pattern `file_path:line_number` to help users navigate:\\n- Example: \\\"The error is handled in src/main.go:45\\\"\\n- Example: \\\"See the implementation in pkg/utils/helper.go:123-145\\\"\\n</code_references>\\n\\n<workflow>\\nFor every task, follow this sequence internally (don't narrate it):\\n\\n**Before acting**:\\n- Search codebase for relevant files\\n- Read files to understand current state\\n- Check memory for stored commands\\n- Identify what needs to change\\n- Use `git log` and `git blame` for additional context when needed\\n\\n**While acting**:\\n- Read entire file before editing it\\n- Before editing: verify exact whitespace and indentation from View output\\n- Use exact text for find/replace (include whitespace)\\n- Make one logical change at a time\\n- After each change: run tests\\n- If tests fail: fix immediately\\n- If edit fails: read more context, don't guess - the text must match exactly\\n- Keep going until query is completely resolved before yielding to user\\n- For longer tasks, send brief progress updates (under 10 words) BUT IMMEDIATELY CONTINUE WORKING - progress updates are not stopping points\\n\\n**Before finishing**:\\n- Verify ENTIRE query is resolved (not just first step)\\n- All described next steps must be completed\\n- Cross-check the original prompt and your own mental checklist; if any feasible part remains undone, continue working instead of responding.\\n- Run lint/typecheck if in memory\\n- Verify all changes work\\n- Keep response under 4 lines\\n\\n**Key behaviors**:\\n- Use find_references before changing shared code\\n- Follow existing patterns (check similar files)\\n- If stuck, try different approach (don't repeat failures)\\n- Make decisions yourself (search first, don't ask)\\n- Fix problems at root cause, not surface-level patches\\n- Don't fix unrelated bugs or broken tests (mention them in final message if relevant)\\n</workflow>\\n\\n<decision_making>\\n**Make decisions autonomously** - don't ask when you can:\\n- Search to find the answer\\n- Read files to see patterns\\n- Check similar code\\n- Infer from context\\n- Try most likely approach\\n- When requirements are underspecified but not obviously dangerous, make the most reasonable assumptions based on project patterns and memory files, briefly state them if needed, and proceed instead of waiting for clarification.\\n\\n**Only stop/ask user if**:\\n- Truly ambiguous business requirement\\n- Multiple valid approaches with big tradeoffs\\n- Could cause data loss\\n- Exhausted all attempts and hit actual blocking errors\\n\\n**When requesting information/access**:\\n- Exhaust all available tools, searches, and reasonable assumptions first.\\n- Never say \\\"Need more info\\\" without detail.\\n- In the same message, list each missing item, why it is required, acceptable substitutes, and what you already attempted.\\n- State exactly what you will do once the information arrives so the user knows the next step.\\n\\nWhen you must stop, first finish all unblocked parts of the request, then clearly report: (a) what you tried, (b) exactly why you are blocked, and (c) the minimal external action required. Don't stop just because one path failed—exhaust multiple plausible approaches first.\\n\\n**Never stop for**:\\n- Task seems too large (break it down)\\n- Multiple files to change (change them)\\n- Concerns about \\\"session limits\\\" (no such limits exist)\\n- Work will take many steps (do all the steps)\\n\\nExamples of autonomous decisions:\\n- File location → search for similar files\\n- Test command → check package.json/memory\\n- Code style → read existing code\\n- Library choice → check what's used\\n- Naming → follow existing names\\n</decision_making>\\n\\n<editing_files>\\n**Available edit tools:**\\n- `edit` - Single find/replace in a file\\n- `multiedit` - Multiple find/replace operations in one file\\n- `write` - Create/overwrite entire file\\n\\nNever use `apply_patch` or similar - those tools don't exist.\\n\\nCritical: ALWAYS read the relevant context of files before editing them in this conversation.\\n\\nWhen using edit tools:\\n1. Read the relevant context first - note the EXACT indentation (spaces vs tabs, count)\\n2. Copy the exact text including ALL whitespace, newlines, and indentation\\n3. Include 3-5 lines of context before and after the target\\n4. Verify your old_string would appear exactly once in the file\\n5. If uncertain about whitespace, include more surrounding context\\n6. Verify edit succeeded\\n7. Run tests\\n\\n**Whitespace matters**:\\n- Count spaces/tabs carefully (use View tool line numbers as reference)\\n- Include blank lines if they exist\\n- Match line endings exactly\\n- When in doubt, include MORE context rather than less\\n\\nEfficiency tips:\\n- Don't re-read files after successful edits (tool will fail if it didn't work)\\n- Same applies for making folders, deleting files, etc.\\n\\nCommon mistakes to avoid:\\n- Editing without reading first\\n- Approximate text matches\\n- Wrong indentation (spaces vs tabs, wrong count)\\n- Missing or extra blank lines\\n- Not enough context (text appears multiple times)\\n- Trimming whitespace that exists in the original\\n- Not testing after changes\\n</editing_files>\\n\\n<whitespace_and_exact_matching>\\nThe Edit tool is extremely literal. \\\"Close enough\\\" will fail.\\n\\n**Before every edit**:\\n1. View the file and locate the exact lines to change\\n2. Copy the text EXACTLY including:\\n - Every space and tab\\n - Every blank line\\n - Opening/closing braces position\\n - Comment formatting\\n3. Include enough surrounding lines (3-5) to make it unique\\n4. Double-check indentation level matches\\n\\n**Common failures**:\\n- `func foo() {` vs `func foo(){` (space before brace)\\n- Tab vs 4 spaces vs 2 spaces\\n- Missing blank line before/after\\n- `// comment` vs `//comment` (space after //)\\n- Different number of spaces in indentation\\n\\n**If edit fails**:\\n- View the file again at the specific location\\n- Copy even more context\\n- Check for tabs vs spaces\\n- Verify line endings\\n- Try including the entire function/block if needed\\n- Never retry with guessed changes - get the exact text first\\n</whitespace_and_exact_matching>\\n\\n<task_completion>\\nEnsure every task is implemented completely, not partially or sketched.\\n\\n1. **Think before acting** (for non-trivial tasks)\\n - Identify all components that need changes (models, logic, routes, config, tests, docs)\\n - Consider edge cases and error paths upfront\\n - Form a mental checklist of requirements before making the first edit\\n - This planning happens internally - don't narrate it to the user\\n\\n2. **Implement end-to-end**\\n - Treat every request as complete work: if adding a feature, wire it fully\\n - Update all affected files (callers, configs, tests, docs)\\n - Don't leave TODOs or \\\"you'll also need to...\\\" - do it yourself\\n - No task is too large - break it down and complete all parts\\n - For multi-part prompts, treat each bullet/question as a checklist item and ensure every item is implemented or answered. Partial completion is not an acceptable final state.\\n\\n3. **Verify before finishing**\\n - Re-read the original request and verify each requirement is met\\n - Check for missing error handling, edge cases, or unwired code\\n - Run tests to confirm the implementation works\\n - Only say \\\"Done\\\" when truly done - never stop mid-task\\n</task_completion>\\n\\n<error_handling>\\nWhen errors occur:\\n1. Read complete error message\\n2. Understand root cause (isolate with debug logs or minimal reproduction if needed)\\n3. Try different approach (don't repeat same action)\\n4. Search for similar code that works\\n5. Make targeted fix\\n6. Test to verify\\n7. For each error, attempt at least two or three distinct remediation strategies (search similar code, adjust commands, narrow or widen scope, change approach) before concluding the problem is externally blocked.\\n\\nCommon errors:\\n- Import/Module → check paths, spelling, what exists\\n- Syntax → check brackets, indentation, typos\\n- Tests fail → read test, see what it expects\\n- File not found → use ls, check exact path\\n\\n**Edit tool \\\"old_string not found\\\"**:\\n- View the file again at the target location\\n- Copy the EXACT text including all whitespace\\n- Include more surrounding context (full function if needed)\\n- Check for tabs vs spaces, extra/missing blank lines\\n- Count indentation spaces carefully\\n- Don't retry with approximate matches - get the exact text\\n</error_handling>\\n\\n<memory_instructions>\\nMemory files store commands, preferences, and codebase info. Update them when you discover:\\n- Build/test/lint commands\\n- Code style preferences\\n- Important codebase patterns\\n- Useful project information\\n</memory_instructions>\\n\\n<code_conventions>\\nBefore writing code:\\n1. Check if library exists (look at imports, package.json)\\n2. Read similar code for patterns\\n3. Match existing style\\n4. Use same libraries/frameworks\\n5. Follow security best practices (never log secrets)\\n6. Don't use one-letter variable names unless requested\\n\\nNever assume libraries are available - verify first.\\n\\n**Ambition vs. precision**:\\n- New projects → be creative and ambitious with implementation\\n- Existing codebases → be surgical and precise, respect surrounding code\\n- Don't change filenames or variables unnecessarily\\n- Don't add formatters/linters/tests to codebases that don't have them\\n</code_conventions>\\n\\n<testing>\\nAfter significant changes:\\n- Start testing as specific as possible to code changed, then broaden to build confidence\\n- Use self-verification: write unit tests, add output logs, or use debug statements to verify your solutions\\n- Run relevant test suite\\n- If tests fail, fix before continuing\\n- Check memory for test commands\\n- Run lint/typecheck if available (on precise targets when possible)\\n- For formatters: iterate max 3 times to get it right; if still failing, present correct solution and note formatting issue\\n- Suggest adding commands to memory if not found\\n- Don't fix unrelated bugs or test failures (not your responsibility)\\n</testing>\\n\\n<tool_usage>\\n- Default to using tools (ls, grep, view, agent, tests, web_fetch, etc.) rather than speculation whenever they can reduce uncertainty or unlock progress, even if it takes multiple tool calls.\\n- Search before assuming\\n- Read files before editing\\n- Always use absolute paths for file operations (editing, reading, writing)\\n- Use Agent tool for complex searches\\n- Run tools in parallel when safe (no dependencies)\\n- When making multiple independent bash calls, send them in a single message with multiple tool calls for parallel execution\\n- Summarize tool output for user (they don't see it)\\n- Never use `curl` through the bash tool it is not allowed use the fetch tool instead.\\n- Only use the tools you know exist.\\n\\n<bash_commands>\\n**CRITICAL**: The `description` parameter is REQUIRED for all bash tool calls. Always provide it.\\n\\nWhen running non-trivial bash commands (especially those that modify the system):\\n- Briefly explain what the command does and why you're running it\\n- This ensures the user understands potentially dangerous operations\\n- Simple read-only commands (ls, cat, etc.) don't need explanation\\n- Use `&` for background processes that won't stop on their own (e.g., `node server.js &`)\\n- Avoid interactive commands - use non-interactive versions (e.g., `npm init -y` not `npm init`)\\n- Combine related commands to save time (e.g., `git status && git diff HEAD && git log -n 3`)\\n</bash_commands>\\n</tool_usage>\\n\\n<proactiveness>\\nBalance autonomy with user intent:\\n- When asked to do something → do it fully (including ALL follow-ups and \\\"next steps\\\")\\n- Never describe what you'll do next - just do it\\n- When the user provides new information or clarification, incorporate it immediately and keep executing instead of stopping with an acknowledgement.\\n- Responding with only a plan, outline, or TODO list (or any other purely verbal response) is failure; you must execute the plan via tools whenever execution is possible.\\n- When asked how to approach → explain first, don't auto-implement\\n- After completing work → stop, don't explain (unless asked)\\n- Don't surprise user with unexpected actions\\n</proactiveness>\\n\\n<final_answers>\\nAdapt verbosity to match the work completed:\\n\\n**Default (under 4 lines)**:\\n- Simple questions or single-file changes\\n- Casual conversation, greetings, acknowledgements\\n- One-word answers when possible\\n\\n**More detail allowed (up to 10-15 lines)**:\\n- Large multi-file changes that need walkthrough\\n- Complex refactoring where rationale adds value\\n- Tasks where understanding the approach is important\\n- When mentioning unrelated bugs/issues found\\n- Suggesting logical next steps user might want\\n- Structure longer answers with Markdown sections and lists, and put all code, commands, and config in fenced code blocks.\\n\\n**What to include in verbose answers**:\\n- Brief summary of what was done and why\\n- Key files/functions changed (with `file:line` references)\\n- Any important decisions or tradeoffs made\\n- Next steps or things user should verify\\n- Issues found but not fixed\\n\\n**What to avoid**:\\n- Don't show full file contents unless explicitly asked\\n- Don't explain how to save files or copy code (user has access to your work)\\n- Don't use \\\"Here's what I did\\\" or \\\"Let me know if...\\\" style preambles/postambles\\n- Keep tone direct and factual, like handing off work to a teammate\\n</final_answers>\\n\\n<env>\\nWorking directory: /tmp/crush-test/TestCoderAgent/glm-5.1/write_tool\\nIs directory a git repo: no\\nPlatform: linux\\nToday's date: 1/1/2025\\n\\n</env>\\n\\n\\n\\n<available_skills>\\n <skill>\\n <name>crush-hooks</name>\\n <description>Use when the user wants to add, write, debug, or configure a Crush hook — gating or blocking tool calls, approving or rewriting tool input before execution, injecting context into tool results, or troubleshooting hook behavior in crush.json.</description>\\n <location>crush://skills/crush-hooks/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n <skill>\\n <name>jq</name>\\n <description>Use when the user needs to query, filter, reshape, extract, create, or construct JSON data — including API responses, config files, log output, or any structured data — or when helping the user write or debug JSON transformations.</description>\\n <location>crush://skills/jq/SKILL.md</location>\\n <type>builtin</type>\\n </skill>\\n</available_skills>\\n\\n<skills_usage>\\nThe `<description>` of each skill is a TRIGGER — it tells you *when* a skill applies. It is NOT a specification of what the skill does or how to do it. The procedure, scripts, commands, references, and required flags live only in the SKILL.md body. You do not know what a skill actually does until you have read its SKILL.md.\\n\\nMANDATORY activation flow:\\n1. Scan `<available_skills>` against the current user task.\\n2. If any skill's `<description>` matches, call the View tool with its `<location>` EXACTLY as shown — before any other tool call that performs the task.\\n3. Read the entire SKILL.md and follow its instructions.\\n4. Only then execute the task, using the skill's prescribed commands/tools.\\n\\nDo NOT skip step 2 because you think you already know how to do the task. Do NOT infer a skill's behavior from its name or description. If you find yourself about to run `bash`, `edit`, or any task-doing tool for a skill-eligible request without having just viewed the SKILL.md, stop and load the skill first.\\n\\nBuiltin skills (type=builtin) use virtual `crush://skills/...` location identifiers. The \\\"crush://\\\" prefix is NOT a URL, network address, or MCP resource — it is a special internal identifier the View tool understands natively. Pass the `<location>` verbatim to View.\\n\\nDo not use MCP tools (including read_mcp_resource) to load skills.\\nIf a skill mentions scripts, references, or assets, they live in the same folder as the skill itself (e.g., scripts/, references/, assets/ subdirectories within the skill's folder).\\n</skills_usage>\\n\\n\\n\\n\",\"role\":\"system\"},{\"content\":\"<system_reminder>This is a reminder that your todo list is currently empty. DO NOT mention this to the user explicitly because they are already aware.\\nIf you are working on tasks that would benefit from a todo list please use the \\\"todos\\\" tool to create one.\\nIf not, please feel free to ignore. Again do not mention this message to the user.</system_reminder>\",\"role\":\"user\"},{\"content\":\"use write to create a new file called config.json with content '{\\\"name\\\": \\\"test\\\", \\\"version\\\": \\\"1.0.0\\\"}'\",\"role\":\"user\"},{\"tool_calls\":[{\"id\":\"call_kLqkWWD0dwlybPAAd0yUiCAx\",\"function\":{\"arguments\":\"{\\\"content\\\": \\\"{\\\\\\\"name\\\\\\\": \\\\\\\"test\\\\\\\", \\\\\\\"version\\\\\\\": \\\\\\\"1.0.0\\\\\\\"}\\\", \\\"file_path\\\": \\\"/tmp/crush-test/TestCoderAgent/glm-5.1/write_tool/config.json\\\"}\",\"name\":\"write\"},\"type\":\"function\"}],\"role\":\"assistant\",\"reasoning_content\":\"Simple request to create a file with write tool.\"},{\"content\":\"<result>\\nFile successfully written: /tmp/crush-test/TestCoderAgent/glm-5.1/write_tool/config.json\\n</result>\",\"tool_call_id\":\"call_kLqkWWD0dwlybPAAd0yUiCAx\",\"role\":\"tool\"}],\"model\":\"glm-5.1\",\"max_tokens\":10000,\"stream_options\":{\"include_usage\":true},\"tool_choice\":\"auto\",\"tools\":[{\"function\":{\"name\":\"bash\",\"strict\":false,\"description\":\"Execute shell commands; long-running commands automatically move to background and return a shell ID.\\n\\n<cross_platform>\\nUses mvdan/sh interpreter (Bash-compatible on all platforms including Windows).\\nUse forward slashes for paths: \\\"ls C:/foo/bar\\\" not \\\"ls C:\\\\foo\\\\bar\\\".\\nCommon shell builtins and core utils available on Windows.\\n</cross_platform>\\n\\n<execution_steps>\\n1. Directory Verification: If creating directories/files, use LS tool to verify parent exists\\n2. Security Check: Banned commands (alias, aria2c, axel, chrome, curl, curlie, firefox, http-prompt, httpie, links, lynx, nc, safari, scp, ssh, telnet, w3m, wget, xh, doas, su, sudo, apk, apt, apt-cache, apt-get, dnf, dpkg, emerge, home-manager, makepkg, opkg, pacman, paru, pkg, pkg_add, pkg_delete, portage, rpm, yay, yum, zypper, at, batch, chkconfig, crontab, fdisk, mkfs, mount, parted, service, systemctl, umount, firewall-cmd, ifconfig, ip, iptables, netstat, pfctl, route, ufw) return error - explain to user. Safe read-only commands execute without prompts\\n3. Command Execution: Execute with proper quoting, capture output\\n4. Auto-Background: Commands exceeding 1 minute (default, configurable via `auto_background_after`) automatically move to background and return shell ID\\n5. Output Processing: Truncate if exceeds 30000 characters\\n6. Return Result: Include errors, metadata with <cwd></cwd> tags\\n</execution_steps>\\n\\n<usage_notes>\\n- Command required, working_dir optional (defaults to current directory)\\n- IMPORTANT: Use Grep/Glob/Agent tools instead of 'find'/'grep'. Use View/LS tools instead of 'cat'/'head'/'tail'/'ls'\\n- Chain with ';' or '&&', avoid newlines except in quoted strings\\n- Each command runs in independent shell (no state persistence between calls)\\n- Prefer absolute paths over 'cd' (use 'cd' only if user explicitly requests)\\n</usage_notes>\\n\\n<background_execution>\\n- Set run_in_background=true to run commands in a separate background shell\\n- Returns a shell ID for managing the background process\\n- Use job_output tool to view current output from background shell\\n- Use job_kill tool to terminate a background shell\\n- IMPORTANT: NEVER use `&` at the end of commands to run in background - use run_in_background parameter instead\\n- Commands that should run in background:\\n * Long-running servers (e.g., `npm start`, `python -m http.server`, `node server.js`)\\n * Watch/monitoring tasks (e.g., `npm run watch`, `tail -f logfile`)\\n * Continuous processes that don't exit on their own\\n * Any command expected to run indefinitely\\n- Commands that should NOT run in background:\\n * Build commands (e.g., `npm run build`, `go build`)\\n * Test suites (e.g., `npm test`, `pytest`)\\n * Git operations\\n * File operations\\n * Short-lived scripts\\n</background_execution>\\n\\n<git_commits>\\nWhen user asks to create git commit:\\n\\n1. Single message with three tool_use blocks (IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - git log (recent commit message style)\\n\\n2. Add relevant untracked files to staging. Don't commit files already modified at conversation start unless relevant.\\n\\n3. Analyze staged changes in <commit_analysis> tags:\\n - List changed/added files, summarize nature (feature/enhancement/bug fix/refactoring/test/docs)\\n - Brainstorm purpose/motivation, assess project impact, check for sensitive info\\n - Don't use tools beyond git context\\n - Draft concise (1-2 sentences) message focusing on \\\"why\\\" not \\\"what\\\"\\n - Use clear language, accurate reflection (\\\"add\\\"=new feature, \\\"update\\\"=enhancement, \\\"fix\\\"=bug fix)\\n - Avoid generic messages, review draft\\n\\n4. Create commit with attribution using HEREDOC:\\n git commit -m \\\"$(cat <<'EOF'\\n Commit message here.\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n\\n Co-Authored-By: Crush <crush@charm.land>\\n\\n\\n EOF\\n )\\\"\\n\\n5. If pre-commit hook fails, retry ONCE. If fails again, hook preventing commit. If succeeds but files modified, MUST amend.\\n\\n6. Run git status to verify.\\n\\nNotes: Use \\\"git commit -am\\\" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.\\n</git_commits>\\n\\n<pull_requests>\\nUse gh command for ALL GitHub tasks. When user asks to create PR:\\n\\n1. Single message with multiple tool_use blocks (VERY IMPORTANT for speed):\\n - git status (untracked files)\\n - git diff (staged/unstaged changes)\\n - Check if branch tracks remote and is up to date\\n - git log and 'git diff main...HEAD' (full commit history from main divergence)\\n\\n2. Create new branch if needed\\n3. Commit changes if needed\\n4. Push to remote with -u flag if needed\\n\\n5. Analyze changes in <pr_analysis> tags:\\n - List commits since diverging from main\\n - Summarize nature of changes\\n - Brainstorm purpose/motivation\\n - Assess project impact\\n - Don't use tools beyond git context\\n - Check for sensitive information\\n - Draft concise (1-2 bullet points) PR summary focusing on \\\"why\\\"\\n - Ensure summary reflects ALL changes since main divergence\\n - Clear, concise language\\n - Accurate reflection of changes and purpose\\n - Avoid generic summaries\\n - Review draft\\n\\n6. Create PR with gh pr create using HEREDOC:\\n gh pr create --title \\\"title\\\" --body \\\"$(cat <<'EOF'\\n\\n ## Summary\\n\\n <1-3 bullet points>\\n\\n ## Test plan\\n\\n [Checklist of TODOs...]\\n\\n\\n \U0001F498 Generated with Crush\\n\\n\\n EOF\\n )\\\"\\n\\nImportant:\\n\\n- Return empty response - user sees gh output\\n- Never update git config\\n</pull_requests>\\n\\n<examples>\\nGood: pytest /foo/bar/tests\\nBad: cd /foo/bar && pytest tests\\n</examples>\\n\",\"parameters\":{\"properties\":{\"auto_background_after\":{\"description\":\"Seconds to wait before automatically moving the command to a background job (default: 60)\",\"type\":\"integer\"},\"command\":{\"description\":\"The command to execute\",\"type\":\"string\"},\"description\":{\"description\":\"A brief description of what the command does, try to keep it under 30 characters or so\",\"type\":\"string\"},\"run_in_background\":{\"description\":\"Set to true (boolean) to run this command in the background. Use job_output to read the output later.\",\"type\":\"boolean\"},\"working_dir\":{\"description\":\"The working directory to execute the command in (defaults to current directory)\",\"type\":\"string\"}},\"required\":[\"description\",\"command\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"download\",\"strict\":false,\"description\":\"Download a URL directly to a local file (binary-safe, streaming, max 600s timeout); overwrites without warning. For reading content into context use fetch.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The local file path where the downloaded content should be saved\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 600)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to download from\",\"type\":\"string\"}},\"required\":[\"url\",\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"edit\",\"strict\":false,\"description\":\"Edit a file by exact find-and-replace; can also create or delete content. For renames/moves use bash. For large edits use write.\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"},\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false)\",\"type\":\"boolean\"}},\"required\":[\"file_path\",\"old_string\",\"new_string\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"multiedit\",\"strict\":false,\"description\":\"Apply multiple find-and-replace edits to a single file in one operation; edits run sequentially. Prefer over edit for multiple changes to the same file. Same exact-match rules as edit apply.\",\"parameters\":{\"properties\":{\"edits\":{\"description\":\"Array of edit operations to perform sequentially on the file\",\"items\":{\"properties\":{\"new_string\":{\"description\":\"The text to replace it with\",\"type\":\"string\"},\"old_string\":{\"description\":\"The text to replace\",\"type\":\"string\"},\"replace_all\":{\"description\":\"Replace all occurrences of old_string (default false).\",\"type\":\"boolean\"}},\"required\":[\"old_string\",\"new_string\"],\"type\":\"object\"},\"type\":\"array\"},\"file_path\":{\"description\":\"The absolute path to the file to modify\",\"type\":\"string\"}},\"required\":[\"file_path\",\"edits\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"fetch\",\"strict\":false,\"description\":\"Fetch raw content from a URL as text, markdown, or html (max 100KB); no AI processing. For analysis or extraction use agentic_fetch.\\n\",\"parameters\":{\"properties\":{\"format\":{\"description\":\"The format to return the content in (text, markdown, or html)\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"},\"url\":{\"description\":\"The URL to fetch content from\",\"type\":\"string\"}},\"required\":[\"url\",\"format\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"glob\",\"strict\":false,\"description\":\"Find files by name/pattern (glob syntax), sorted by modification time; max 100 results; skips hidden files. Use grep to search file contents.\\n\",\"parameters\":{\"properties\":{\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The glob pattern to match files against\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"grep\",\"strict\":false,\"description\":\"Search file contents by regex or literal text; returns matching file paths sorted by modification time (max 100); respects .gitignore. Use glob to filter by filename, not contents.\\n\",\"parameters\":{\"properties\":{\"include\":{\"description\":\"File pattern to include in the search (e.g. \\\"*.js\\\", \\\"*.{ts,tsx}\\\")\",\"type\":\"string\"},\"literal_text\":{\"description\":\"If true, the pattern will be treated as literal text with special regex characters escaped. Default is false.\",\"type\":\"boolean\"},\"path\":{\"description\":\"The directory to search in. Defaults to the current working directory.\",\"type\":\"string\"},\"pattern\":{\"description\":\"The regex pattern to search for in file contents\",\"type\":\"string\"}},\"required\":[\"pattern\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"ls\",\"strict\":false,\"description\":\"List files and directories as a tree; skips hidden files and common system dirs; max 1000 files. Use glob to find files by pattern, grep to search contents.\\n\",\"parameters\":{\"properties\":{\"depth\":{\"description\":\"The maximum depth to traverse\",\"type\":\"integer\"},\"ignore\":{\"description\":\"List of glob patterns to ignore\",\"items\":{\"type\":\"string\"},\"type\":\"array\"},\"path\":{\"description\":\"The path to the directory to list (defaults to current working directory)\",\"type\":\"string\"}},\"required\":[],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"sourcegraph\",\"strict\":false,\"description\":\"Search code across public GitHub repositories via Sourcegraph; supports regex, language/repo/file filters, and symbol search (max 20 results). Only searches public repos.\\n\",\"parameters\":{\"properties\":{\"context_window\":{\"description\":\"The context around the match to return (default: 10 lines)\",\"type\":\"integer\"},\"count\":{\"description\":\"Optional number of results to return (default: 10, max: 20)\",\"type\":\"integer\"},\"query\":{\"description\":\"The Sourcegraph search query\",\"type\":\"string\"},\"timeout\":{\"description\":\"Optional timeout in seconds (max 120)\",\"type\":\"integer\"}},\"required\":[\"query\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"view\",\"strict\":false,\"description\":\"Read a file by path with line numbers; supports offset and line limit (default 200, max 200KB returned file content section); renders images (PNG, JPEG, GIF, WebP); use ls for directories.\\n\",\"parameters\":{\"properties\":{\"file_path\":{\"description\":\"The path to the file to read\",\"type\":\"string\"},\"limit\":{\"description\":\"The number of lines to read (defaults to 200)\",\"type\":\"integer\"},\"offset\":{\"description\":\"The line number to start reading from (0-based)\",\"type\":\"integer\"}},\"required\":[\"file_path\"],\"type\":\"object\"}},\"type\":\"function\"},{\"function\":{\"name\":\"write\",\"strict\":false,\"description\":\"Create or overwrite a file with given content; auto-creates parent dirs. Cannot append. Read the file first to avoid conflicts. For surgical changes use edit or multiedit.\",\"parameters\":{\"properties\":{\"content\":{\"description\":\"The content to write to the file\",\"type\":\"string\"},\"file_path\":{\"description\":\"The path to the file to write\",\"type\":\"string\"}},\"required\":[\"file_path\",\"content\"],\"type\":\"object\"}},\"type\":\"function\"}],\"stream\":true}"

340 headers:

394 headers:

341 Accept:

395 Accept:

342 - application/json

396 - application/json

@@ -353,15 +407,21 @@ interactions:

353 content_length: -1

407 content_length: -1

354 uncompressed: true

408 uncompressed: true

355 body: |+

409 body: |+

356− data: {"id":"chatcmpl-e0f83d2e7efaff504d6bfede","object":"chat.completion.chunk","created":1779218937,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

410+ data: {"id":"chatcmpl-c5f5fee46ba6c248fb29dc2e","object":"chat.completion.chunk","created":1780069706,"model":"glm-5.1","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

411+

412+ data: {"id":"chatcmpl-c5f5fee46ba6c248fb29dc2e","object":"chat.completion.chunk","created":1780069706,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Created"},"finish_reason":null}]}

413+

414+ data: {"id":"chatcmpl-c5f5fee46ba6c248fb29dc2e","object":"chat.completion.chunk","created":1780069706,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":" `"},"finish_reason":null}]}

415+

416+ data: {"id":"chatcmpl-c5f5fee46ba6c248fb29dc2e","object":"chat.completion.chunk","created":1780069706,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"config"},"finish_reason":null}]}

357

417

358− data: {"id":"chatcmpl-e0f83d2e7efaff504d6bfede","object":"chat.completion.chunk","created":1779218937,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"Done"},"finish_reason":null}]}

418+ data: {"id":"chatcmpl-c5f5fee46ba6c248fb29dc2e","object":"chat.completion.chunk","created":1780069706,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":".json"},"finish_reason":null}]}

359

419

360− data: {"id":"chatcmpl-e0f83d2e7efaff504d6bfede","object":"chat.completion.chunk","created":1779218937,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"."},"finish_reason":null}]}

420+ data: {"id":"chatcmpl-c5f5fee46ba6c248fb29dc2e","object":"chat.completion.chunk","created":1780069706,"model":"glm-5.1","choices":[{"index":0,"delta":{"content":"`."},"finish_reason":null}]}

361

421

362− data: {"id":"chatcmpl-e0f83d2e7efaff504d6bfede","object":"chat.completion.chunk","created":1779218937,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

422+ data: {"id":"chatcmpl-c5f5fee46ba6c248fb29dc2e","object":"chat.completion.chunk","created":1780069706,"model":"glm-5.1","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

363

423

364− data: {"id":"chatcmpl-e0f83d2e7efaff504d6bfede","object":"chat.completion.chunk","created":1779218937,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":107,"completion_tokens":4,"total_tokens":7919,"prompt_tokens_details":{"cached_tokens":7808},"completion_tokens_details":{"reasoning_tokens":3},"cost":{"usd":0.0001913,"hypercredits":0.0038259999999999995},"remaining":{"hypercredits":0}}}

424+ data: {"id":"chatcmpl-c5f5fee46ba6c248fb29dc2e","object":"chat.completion.chunk","created":1780069706,"model":"glm-5.1","choices":[],"usage":{"prompt_tokens":102,"completion_tokens":7,"total_tokens":7917,"prompt_tokens_details":{"cached_tokens":7808},"cost":{"usd":0.00016302000000000002,"hypercredits":0.0032604},"remaining":{"hypercredits":0}}}

365

425

366 data: [DONE]

426 data: [DONE]

367

427

@@ -370,4 +430,4 @@ interactions:

370 - text/event-stream

430 - text/event-stream

371 status: 200 OK

431 status: 200 OK

372 code: 200

432 code: 200

373− duration: 1.999880834s

433+ duration: 628.357791ms

internal/agent/tools/bash.md.tpl+1−1

@@ -85,7 +85,7 @@ When user asks to create git commit:

866. Run git status to verify.

88−Notes: Use "git commit -am" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m or GIT_EDITOR=true.

88+Notes: Use "git commit -am" when possible, don't stage unrelated files, NEVER update config, don't push, no -i flags, no empty commits, return empty response, when rebasing always use -m.

89</git_commits>

91<pull_requests>

internal/shell/run.go+41

@@ -89,6 +89,7 @@ func Run(ctx context.Context, opts RunOptions) (err error) {

89// Crush handler stack. Shared by the stateless [Run] entrypoint and the

90// stateful [Shell] so the two surfaces cannot drift.

91func newRunner(cwd string, env []string, stdin io.Reader, stdout, stderr io.Writer, blockFuncs []BlockFunc) (*interp.Runner, error) {

92+ env = withNonInteractiveEnv(env)

92 return interp.New(

93 return interp.New(

93 interp.StdIO(stdin, stdout, stderr),

94 interp.StdIO(stdin, stdout, stderr),

94 interp.Interactive(false),

95 interp.Interactive(false),

@@ -98,6 +99,46 @@ func newRunner(cwd string, env []string, stdin io.Reader, stdout, stderr io.Writ

98 )

99 )

99}

100}

100

101

102+// nonInteractiveEnvVars are forced on every shell execution to prevent

103+// commands from hanging on a nonexistent TTY. These are always applied

104+// regardless of the caller's environment because Crush shells are never

105+// interactive — preserving user preferences like EDITOR=nvim only causes

106+// hangs, not useful behavior.

107+var nonInteractiveEnvVars = []string{

108+ "TERM=xterm-256color",

109+ "GIT_EDITOR=false",

110+ "EDITOR=false",

111+ "VISUAL=false",

112+ "JJ_EDITOR=false",

113+ "JJ_PAGER=cat",

114+ "GIT_PAGER=cat",

115+ "PAGER=cat",

116+}

117+

118+// withNonInteractiveEnv returns env with nonInteractiveEnvVars forced in,

119+// replacing any existing values for those keys. The returned slice is a

120+// new allocation safe to use concurrently with the input.

121+func withNonInteractiveEnv(env []string) []string {

122+ // Build a set of override keys for fast lookup.

123+ overrideKeys := make(map[string]bool, len(nonInteractiveEnvVars))

124+ for _, kv := range nonInteractiveEnvVars {

125+ if key, _, ok := strings.Cut(kv, "="); ok {

126+ overrideKeys[key] = true

127+ }

128+ }

129+

130+ // Copy env, filtering out any keys we will override.

131+ result := make([]string, 0, len(env)+len(nonInteractiveEnvVars))

132+ for _, e := range env {

133+ if key, _, ok := strings.Cut(e, "="); ok && overrideKeys[key] {

134+ continue

135+ }

136+ result = append(result, e)

137+ }

138+

139+ return append(result, nonInteractiveEnvVars...)

140+}

141+

101// standardHandlers returns the exec-handler middleware chain used by both

142// standardHandlers returns the exec-handler middleware chain used by both

102// [Run] and [Shell]. Order matters:

143// [Run] and [Shell]. Order matters:

103// 1. builtins first (so Crush's in-process jq wins over any PATH binary);

144// 1. builtins first (so Crush's in-process jq wins over any PATH binary);

internal/shell/run_test.go+69

@@ -5,6 +5,7 @@ import (

5 "context"

6 "errors"

7 "fmt"

8+ "slices"

8 "strings"

9 "strings"

9 "sync"

10 "sync"

10 "testing"

11 "testing"

@@ -240,6 +241,74 @@ func TestRun_RequiresCwd(t *testing.T) {

240 }

241 }

241}

242}

242

243

244+func TestWithNonInteractiveEnv_Empty(t *testing.T) {

245+ t.Parallel()

246+ result := withNonInteractiveEnv(nil)

247+ // All defaults must be present.

248+ for _, want := range nonInteractiveEnvVars {

249+ if !slices.Contains(result, want) {

250+ t.Errorf("missing default %q in result", want)

251+ }

252+ }

253+}

254+

255+func TestWithNonInteractiveEnv_OverridesExisting(t *testing.T) {

256+ t.Parallel()

257+ env := []string{"EDITOR=nvim", "PAGER=less", "FOO=bar"}

258+ result := withNonInteractiveEnv(env)

259+

260+ // EDITOR and PAGER must be overridden, not preserved.

261+ for _, e := range result {

262+ if e == "EDITOR=nvim" {

263+ t.Error("EDITOR=nvim should have been overridden")

264+ }

265+ if e == "PAGER=less" {

266+ t.Error("PAGER=less should have been overridden")

267+ }

268+ }

269+ // FOO must survive.

270+ if !slices.Contains(result, "FOO=bar") {

271+ t.Error("FOO=bar should be preserved")

272+ }

273+}

274+

275+func TestWithNonInteractiveEnv_NoPrefixCollision(t *testing.T) {

276+ t.Parallel()

277+ // EDITORIAL should NOT match EDITOR.

278+ env := []string{"EDITORIAL=yes", "GITHUB_TOKEN=secret"}

279+ result := withNonInteractiveEnv(env)

280+

281+ foundEditorial := false

282+ foundGithub := false

283+ for _, e := range result {

284+ if e == "EDITORIAL=yes" {

285+ foundEditorial = true

286+ }

287+ if e == "GITHUB_TOKEN=secret" {

288+ foundGithub = true

289+ }

290+ }

291+ if !foundEditorial {

292+ t.Error("EDITORIAL=yes should not be removed by EDITOR override")

293+ }

294+ if !foundGithub {

295+ t.Error("GITHUB_TOKEN=secret should not be removed")

296+ }

297+}

298+

299+func TestWithNonInteractiveEnv_SliceIndependence(t *testing.T) {

300+ t.Parallel()

301+ env := []string{"FOO=bar"}

302+ result := withNonInteractiveEnv(env)

303+ // Mutating the input must not affect the result.

304+ env[0] = "FOO=baz"

305+ for _, e := range result {

306+ if e == "FOO=baz" {

307+ t.Error("result shares backing array with input")

308+ }

309+ }

310+}

311+

243func TestRun_DiscardsNilWriters(t *testing.T) {

312func TestRun_DiscardsNilWriters(t *testing.T) {

244 // No panic when Stdout/Stderr are nil.

313 // No panic when Stdout/Stderr are nil.

245 err := Run(t.Context(), RunOptions{

314 err := Run(t.Context(), RunOptions{