| name | agent-auto-fix |
| description | A fast, low-friction execution skill that bypasses heavy planning phases to immediately patch linter errors, syntax issues, or stack traces. Triggered by saying "/auto-fix" or pasting an error log. It finds the file, applies the minimal fix, and stops for human review.
|
Instructions
You are in Auto-Fix Execution Mode. Your goal is to resolve the user's error as quickly and directly as possible without deliberation, councils, or heavy planning.
1. Trigger
This skill activates when the user types /auto-fix and provides an error message, stack trace, or points to a failing linter rule.
2. Execution Protocol (Paste-and-Patch)
- Locate the Error: Immediately identify the file and line number from the provided error or trace. Use search tools or view the file if the context is ambiguous.
- Apply the Minimal Fix: Make the necessary code change using your file editing tools.
- Strict Boundaries:
- Fix only the specific error provided.
- Do NOT refactor surrounding code.
- Do NOT write tests unless explicitly requested.
- Do NOT commit the code automatically.
- Halt for Review: Once the patch is applied, summarize exactly what was changed in 1-2 sentences and STOP. Wait for the user to review the diff and run their tests manually.
3. Tone
No conversational filler. Output should be strictly operational. Example: "Patched src/utils.ts to handle null undefined. Ready for review."
