AgentSkillsCN

debug

系统性地诊断并修复Bug、CI失败以及测试失败。可通过`/debug`、调试问题、测试失败、CI错误等指令触发。

SKILL.md
--- frontmatter
name: debug
description: >
  Systematically diagnose and fix bugs, CI failures, and test failures.
  Triggers: /debug, debugging issues, test failures, CI errors
allowed-tools: Bash, Read, Glob, Grep, Edit
argument-hint: "[beads-id|error-description]"

Debug Skill

Systematically diagnose and fix bugs, CI failures, and test failures. Integrates with beads for issue tracking.

Argument Parsing

Parse $ARGUMENTS:

  • •Beads issue ID (e.g., claude-xyz) → debug that specific issue
  • •Error message or description → debug that problem
  • •No args → check for failing tests/CI on current branch

Workflow

1. Issue Setup

If beads issue provided:

bash
bd show <id>
bd update <id> --status in_progress

If no beads issue:

bash
bd create "Debug: <problem>" --type bug --priority 1 --description "## Steps to Reproduce
- <observed symptoms and error output>

## Acceptance Criteria
- Bug is fixed and verified by passing tests"

Validate: bd lint <id> — if it fails, bd edit <id> --description to fix violations.

2. Gather Diagnostics (Parallel)

Run these in parallel:

bash
git branch --show-current
git log --oneline -5
git diff --stat
gh pr checks  # or gh run list --limit 3

If test failure mentioned, run failing tests to reproduce.

3. Investigate Systematically

  1. •Read error output carefully
  2. •Trace to root cause (don't guess)
  3. •Read relevant source files
  4. •Check recent changes that may have introduced the bug
  5. •Use Grep/Glob to find related code

4. Fix the Issue

  1. •Make minimal, targeted changes
  2. •Re-run failing tests/checks to verify fix
  3. •If fix works: bd close <id>
  4. •If fix doesn't work: bd update <id> --notes "Findings: ..."

5. Report

Output format:

code
## Problem
- [what was wrong]

## Root Cause
- [traced cause]

## Fix Applied
- [minimal changes made]

## Verification
- [test results / CI status]

Style Rules

  • •Keep concise — bullet points, not prose
  • •No emoji
  • •Steps numbered and actionable
  • •Prefer reading error output over guessing
  • •Make minimal fixes — don't refactor surrounding code
  • •Use parallel tool calls when gathering independent data