Debug Methodology — AI skill for Claude Code
Systematic debugging methodology for AI agents and developers.
How to install Debug Methodology
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open abczsl520/debug-methodology and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Debug Methodology does
Systematic debugging methodology for AI agents and developers. Prevents common anti-patterns like patch-chaining and wrong-environment restarts.
Alternatives in AI
- Claude Cookbook — Official notebooks and recipes for common patterns (RAG, tool use, Skills, MCP) 35.3k ★
- Dive Into Claude Code — A Systematic Analysis and Discussion of Claude Code for Designing Today's and Future AI Agent Systems 2.1k ★
- Gsap Skills — Official AI skills for GSAP 1.7k ★
README
🔍 Debug Methodology
[](https://clawhub.com) [](LICENSE) [](https://github.com/openclaw/openclaw)
**系统化调试方法论** — 适用于 AI Agent 和开发者的通用调试规范。
从真实生产事故中提炼,结合 Nicole Tietz、Brendan Gregg、Julia Evans 等业界顶级工程师的方法论。
为什么需要这个?
AI Agent 在调试时容易陷入以下陷阱:
- 🚨 醉汉反模式 — 随机改代码直到问题消失
- 🚨 路灯反模式 — 只在熟悉的地方找,而不是问题真正在的地方
- 🚨 补丁链 — 每次报新错就修那个错,越改越乱
- 🚨 忽略用户 — 用户说"改了X就坏了",却继续自己猜
这套方法论提供了一个**强制性的调试流程**,避免这些常见错误。
核心流程
Phase 1: STOP → 动手前先搞清现状(进程、环境、启动命令)
Phase 2: THINK → 形成一个假设(优先查自己改了什么)
Phase 3: TEST → 一次改一个,验证后再继续
Phase 4: DETECT → 改了2次没好?全部回退,重新来
快速决策树
Error appears
├─ Was I just editing? → DIFF my changes → REVERT if suspect
├─ Service won't start? → CHECK startup command + environment
├─ New error after fix? → STOP (patch chain!) → Revert all → Phase 1
├─ User reports regression? → DIFF before/after their last known-good
└─ Intermittent? → CHECK logs + external dependencies + timing
作为 OpenClaw Skill 使用
将 `SKILL.md` 放入你的 skills 目录:
mkdir -p ~/.agents/skills/debug-methodology
cp SKILL.md ~/.agents/skills/debug-methodology/
重启 OpenClaw 后,所有 session 遇到调试场景会自动加载这套方法论。
完整规范
详见 [SKILL.md](SKILL.md) — 包含:
- 4阶段调试流程(STOP → Hypothesize → Test → Patch-Chain Detection)
- 4大反模式警告(醉汉/路灯/货物崇拜/忽略用户)
- 环境检查清单(runtime/deps/config/process manager/logs/backup)
- 部署安全流程(拉取→备份→修改→测试→部署→验证)
- 快速决策树
起源
这套方法论源自一次真实的生产事故:
修复一个简单的超时问题(2步就能搞定),却因为重启服务时没用虚拟环境,走了10步弯路。事后复盘发现,如果一开始跑一句 `ps -p -o command=` 就能避免所有问题。
由此总结出这套通用调试规范,并结合业界最佳实践形成了完整的方法论。
Install
clawhub install debug-methodology
Wiki
更详细的案例分析和扩展内容请查看 [Wiki](../../wiki)。
🔗 Part of the AI Dev Quality Suite
| Skill | Purpose | Install | |-------
Related Skills
Bug Report Agent Skills
AI agent skill for automated bug report processing and systematic debugging. Enforces Detective Methodology wo
Vibe Tuning
Stop fixing the same AI mistake twice. Methodology + Claude Code skill that turns wrong AI results into perman
Contextkit
by Cihat Gündüz - A systematic development framework that transforms Claude Code into a proactive development
Ring
89 skills and 38 specialized agents that enforce proven engineering practices for AI-assisted development. TDD
Superpowers Multi AI
⚡ Modular prompt-engineering framework for AI Agents (Cursor, Claude Code, Copilot). Forces structured SDLC, l
No No Debug
No-No Debug — Self-evolution system for AI coding assistants. 10 minutes writing code, 2 hours debugging? This
Related Agents
Platform Code Reviewer
Use this agent for reviewing Go, Terraform, and GCP infrastructure code with a focus on semantic correctness,
Node Passenger Engineer
Node.js applications running under cPanel's "Setup Node.js App" (Phusion Passenger) - app entrypoints, the nod
Shared Package Agent
Conditionally-invoked pre-implementation gate for changes touching shared/, Common/, or Kernel/ directories —