AgentSkillsCN

debug-loop

支持最多5次修复尝试的自主调试循环

SKILL.md
--- frontmatter
name: debug-loop
description: Autonomous debugging loop with up to 5 fix attempts
args: <error description or stack trace>
allowed-tools: Bash, Read, Edit, Write, Glob, Grep, Task

Issue to debug

$ARGUMENTS

Your task

Debug this issue autonomously. Follow this loop:

For each iteration (max 5 attempts):

  1. •Reproduce: Run relevant tests or reproduction steps to confirm the failure
  2. •Analyze: Read stack traces, error messages, relevant code paths
  3. •Hypothesize: Form a specific theory about the root cause
  4. •Fix: Implement the fix
  5. •Verify: Re-run tests/reproduction

Logging

After each iteration, output a summary:

code
## Attempt N

**Tried**: [specific change made]
**Result**: [pass/fail + details]
**Next**: [what you'll try if this failed, or "RESOLVED" if fixed]

Exit conditions

  • •Success: Tests pass. Summarize the root cause and fix.
  • •5 attempts exhausted: Summarize all approaches tried and ask for guidance.
  • •Truly ambiguous: If you need clarification on expected behavior or the issue is unclear, ask. Otherwise keep iterating.

Rules

  • •Start by finding and running the relevant test file(s)
  • •If no tests exist, create a minimal reproduction first
  • •Each attempt should try a meaningfully different approach
  • •Check environment variables and secrets early for auth errors (401, 403)
  • •Read error messages carefully; the answer is often in the stack trace