A 4-phase framework for systematic root cause analysis and regression prevention.
When to Use
When the user reports a bug that isn't immediately obvious
When a fix didn't work or the bug keeps coming back
When multiple symptoms might share a root cause
Process
Phase 1: Reproduce
Establish exact reproduction steps
Identify the minimal reproduction case
Document expected vs actual behavior
Note environment details (OS, versions, config)
Phase 2: Isolate
Binary search the problem space:
Does it happen with minimal input?
Does it happen in a clean environment?
When did it start? (git bisect if applicable)
Narrow to the specific module/function/line
Rule out environment, data, and timing issues
Phase 3: Root Cause
Identify the actual cause (not just the symptom)
Ask "why?" until you reach the true root:
Why did the error occur? → Because X was null
Why was X null? → Because the API returned early
Why did it return early? → Because validation failed silently
Root cause: Silent validation failure with no error propagation
Verify the root cause explains ALL observed symptoms
Phase 4: Prevent
Fix the root cause (not just the symptom)
Add a test that would have caught this
Check for the same pattern elsewhere in the codebase
Document what was learned (if it could bite again)
Rules
Never skip Phase 1 — if you can't reproduce it, you can't verify the fix
A fix that doesn't address root cause will regress
If Phase 2 takes too long, check assumptions from Phase 1
Always verify the fix in the same way you reproduced the bug
1---2name: lobster-debugging3description: Lobster Debugging4---5# Lobster Debugging67A 4-phase framework for systematic root cause analysis and regression prevention.89## When to Use1011- When the user reports a bug that isn't immediately obvious12- When a fix didn't work or the bug keeps coming back13- When multiple symptoms might share a root cause1415## Process1617### Phase 1: Reproduce18- Establish exact reproduction steps19- Identify the minimal reproduction case20- Document expected vs actual behavior21- Note environment details (OS, versions, config)2223### Phase 2: Isolate24- Binary search the problem space:25 - Does it happen with minimal input?26 - Does it happen in a clean environment?27 - When did it start? (git bisect if applicable)28- Narrow to the specific module/function/line29- Rule out environment, data, and timing issues3031### Phase 3: Root Cause32- Identify the actual cause (not just the symptom)33- Ask "why?" until you reach the true root:34 - Why did the error occur? → Because X was null35 - Why was X null? → Because the API returned early36 - Why did it return early? → Because validation failed silently37 - **Root cause**: Silent validation failure with no error propagation38- Verify the root cause explains ALL observed symptoms3940### Phase 4: Prevent41- Fix the root cause (not just the symptom)42- Add a test that would have caught this43- Check for the same pattern elsewhere in the codebase44- Document what was learned (if it could bite again)4546### Rules47- Never skip Phase 1 — if you can't reproduce it, you can't verify the fix48- A fix that doesn't address root cause will regress49- If Phase 2 takes too long, check assumptions from Phase 150- Always verify the fix in the same way you reproduced the bug
Run npx skillmds@latest add comcclelland/lobster-debugging in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.
Lobster Debugging It is listed under Coding & Dev Tools on SkillMD.
This skill has not completed SkillMD's automated safety review yet. Independent scanners report: SkillSpector: PASS, Skill Scanner: PASS. Capability flags: docs only. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.
This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.
Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.
comcclelland (@comcclelland) published this skill. Their other Agent Skills are listed on their SkillMD profile.