360 Secure SkillHub
360 安全技能中心

Lobster Debugging

面向软件缺陷排查的调试方法论 Skill,按根因调查、条件等待、纵深防御、验证四阶段执行,强调先定位根因再修复、用测试和防护防止复发,并在两次修复失败后回到调查阶段。

wangxiaofei860208-source
wangxiaofei860208-source
热度 49.19
最近更新 2026-05-11 17:40:03
分类:开发依赖:代码执行 / 读写文件应用场景:调试排错 / 工作流

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add lobster-debugging
Metadata
name
lobster-debugging
version
1.0.0
description
Systematic debugging framework. 4-phase root-cause analysis with defense-in-depth. Never guess, never patch symptoms. From Claude Code + Superpowers.
author
Super Lobster 🦞
tags
["debugging","testing","quality","root-cause"]
license
MIT

Lobster Debugging 🦞

Systematic root-cause debugging. No guessing. No symptom patches. Ever.

The Iron Law

NO FIXES WITHOUT ROOT CAUSE INVESTIGATION FIRST

4-Phase Process

Phase 1: Root Cause Investigation

  1. Read error messages completely (they often contain the solution)
  2. Reproduce consistently (can't fix what you can't reproduce)
  3. Check recent changes (git diff, recent commits)
  4. Add diagnostic instrumentation at component boundaries
  5. Binary search the codebase to isolate the failure point

Phase 2: Condition-Based Waiting

  • Replace sleep(5000) with event-based waiting
  • Wait for conditions, not timeouts
  • Flaky tests = unfixed bug, not "timing issue"

Phase 3: Defense-in-Depth

  • Fix the root cause
  • Add a test that would have caught it
  • Add a guard that prevents the class of bug
  • Document why it happened

Phase 4: Academic Verification

  • Prove the fix works with a test
  • Prove the fix doesn't break anything else
  • Prove the fix handles edge cases
  • Prove the original bug can't recur

Red Flags — Stop and Re-Investigate

  • "Just one quick fix" — there's no such thing
  • Already tried multiple fixes — you don't understand the problem
  • Under time pressure — rushing guarantees rework
  • "Seems simple" — simple bugs have root causes too

When to Escalate

After 2 failed fix attempts:

  1. Stop. You're guessing.
  2. Go back to Phase 1.
  3. If still stuck, bring in a second agent with fresh context.