parallel-bugfix
Fix bugs using 3 parallel agents with different strategies, pick the winning approach
- License 许可
- MIT
- 最近更新
- 2026-10-02
- Stars
- ★ 9
- Forks
- 1
01 前言 Front Matter
- name
- parallel-bugfix
- description
- Fix bugs using 3 parallel agents with different strategies, pick the winning approach
02 资源文件 2
- 参考文档evals/evals.json6.8 KB
- 参考文档references/ai-image-prompting.md7.9 KB
03 SKILL.md 正文
Parallel Bug-Fix with Test Validation
Usage
Invoke with: /parallel-bugfix [bug description]
Workflow
- Analyze the bug — identify the suspected root cause file(s) and symptoms
- Spawn 3 parallel sub-agents using the Task tool, each in a worktree (
isolation: "worktree"):
Agent 1: Minimal Targeted Fix
- Fix only the suspected root cause with the smallest possible change
- Run build/lint after edit
- Report pass/fail + diff size
Agent 2: Refactor to Eliminate Bug Class
- Refactor the surrounding code to make the bug class impossible
- Run build/lint after edit
- Report pass/fail + diff size
Agent 3: Component Rewrite (if <200 lines)
- Rewrite the affected component from scratch
- If component is >200 lines, skip this agent and report "too large to rewrite"
- Run build/lint after edit
- Report pass/fail + diff size
- Compare results from all 3 agents:
- Pick the approach that passes all checks with the smallest diff
- If none pass, synthesize the best elements into a final fix
- Apply the winning fix to the main working tree
- Run final validation — build + lint + any existing tests
Rules
- Each agent gets MAX 200 lines of code changes
- Always run build verification before declaring success
- If all 3 fail, report findings and ask user for direction
04 同仓库更多 Skills
writing-skills
Use when creating new skills, editing existing skills, or verifying skills work before deployment
writing-plans
Use when you have a spec or requirements for a multi-step task, before touching code
docs:write-concisely
Apply writing rules to any documentation that humans will read. Makes your writing clearer, stronger, and more professional.
kaizen:why
Iterative Five Whys root cause analysis drilling from symptoms to fundamentals
verification-loop
A comprehensive verification system for Claude Code sessions.
verification-before-completion
Use when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always