AgentSkillsCN

Debugging

调试

SKILL.md
--- frontmatter
name: "Debugging"
id: "debugging"
version: "1.0"
claude_subagent: "debugging-assistant"
fallback: "self-check"
min_tad_version: "2.1"
platforms: ["claude", "codex", "gemini"]

Debugging Skill

Purpose

Systematically diagnose and fix bugs, errors, and unexpected behavior in code through structured analysis.

When to Use

  • •During Gate 3 (implementation quality)
  • •When encountering runtime errors
  • •When tests fail unexpectedly
  • •For logic error investigation
  • •For performance issue root cause analysis

Checklist

Critical (P0) - Must Pass

  • • Root cause identified (not just symptoms)
  • • Fix addresses root cause
  • • No regression introduced
  • • Error no longer reproducible
  • • Related code paths checked

Important (P1) - Should Pass

  • • Test added to prevent regression
  • • Error handling improved
  • • Logging added for future debugging
  • • Documentation updated if behavior changed
  • • Similar patterns checked elsewhere

Nice-to-have (P2) - Informational

  • • Debug notes documented
  • • Monitoring/alerts added
  • • Performance impact assessed
  • • Related improvements identified
  • • Knowledge shared with team

Suggestions (P3) - Optional

  • • Automated prevention possible
  • • Tool/process improvement ideas
  • • Training opportunity identified
  • • Architecture improvement suggested

Pass Criteria

LevelRequirement
P0All items pass
P1Max 1 failure
P2Informational
P3Optional

Evidence Output

Path: .tad/evidence/reviews/{date}-debugging-{task}.md

Execution Contract

  • •Input: error_description, stack_trace, file_paths[], context{}
  • •Output: {passed: bool, root_cause: string, fix_applied: bool, evidence_path: string}
  • •Timeout: 300s
  • •Parallelizable: false

Claude Enhancement

When running on Claude Code, call subagent debugging-assistant for deeper analysis. Reference: .tad/templates/output-formats/debugging-format.md

Debugging Process

1. Reproduce

  • • Error consistently reproducible
  • • Minimal reproduction case identified
  • • Environment factors documented
  • • Trigger conditions understood

2. Isolate

  • • Scope narrowed to specific module
  • • Related components ruled out
  • • Data dependencies identified
  • • Timing/order dependencies checked

3. Analyze

  • • Stack trace examined
  • • Logs reviewed
  • • State inspection performed
  • • Hypothesis formed

4. Fix

  • • Root cause addressed
  • • Minimal change principle followed
  • • Side effects considered
  • • Rollback plan available

5. Verify

  • • Original error resolved
  • • No new errors introduced
  • • Edge cases tested
  • • Performance acceptable

6. Document

  • • Fix documented
  • • Lessons learned captured
  • • Prevention measures identified
  • • Knowledge shared