Investigate

Investigate — Root Cause Analysis

Investigation is the discipline of understanding a problem before fixing it. This skill enforces a strict protocol: no fixes until the root cause is found.

Enforced: while this skill is active, Edit, Write, and NotebookEdit are removed via disallowed-tools. The no-fixes rule is a hard constraint here, not a request. Once you have found and stated the root cause, conclude the investigation — the fix happens as a separate step outside this skill.

Announce at start: "I'm using the investigate skill — no fixes until we find the root cause."

The Iron Law

┌──────────────────────────────────────────────┐
│  NO FIXES WITHOUT ROOT CAUSE INVESTIGATION   │
│                                              │
│  If you haven't found the root cause,        │
│  you cannot propose a fix.                   │
└──────────────────────────────────────────────┘

This is not a suggestion. This is a hard constraint. Guessing causes more bugs than it fixes.

Process

Phase 1: Scope Lock

Before investigating, lock the investigation scope to prevent it from sprawling:

1. State the symptom — What exactly is broken? Be precise.
2. Identify the module — Which part of the codebase is affected?
3. Freeze to module — Investigation stays within this module until evidence points elsewhere.

Example:

Symptom: API returns 500 on POST /api/webhooks
Module: packages/api/src/routes/webhooks.ts
Freeze: Investigation limited to webhook handler + its direct dependencies

Why freeze? Without scope, investigation becomes exploration. Exploration finds interesting things but doesn't fix bugs.

Phase 2: Evidence Collection

Gather evidence BEFORE forming any hypothesis:

1. Read the error — Full stack trace, error message, error code. Not a glance — read every line.

2. Reproduce — Can you trigger it reliably? What are the exact steps?

   • If reproducible: proceed
   • If intermittent: gather more data points, don't guess

3. Check the timeline — What changed recently?

   git log --oneline -20
   git diff HEAD~3
   

4. Trace the data flow — Follow the data from input to error:

   • What value enters the function?
   • What transformation happens?
   • Where does it break?
   • Trace BACKWARD from the error to the source

5. Check boundaries — For multi-component systems, verify data at each boundary:

   • API → service: is the request correct?
   • Service → database: is the query correct?
   • Database → response: is the result expected?

Phase 3: Hypothesis

Form ONE hypothesis based on evidence:

"I think [X] is the root cause because [evidence Y shows Z]"

Requirements:

• Must be specific (not "something is wrong with auth")
• Must be supported by evidence collected in Phase 2
• Must be testable with a single, minimal change

Phase 4: Test

Test the hypothesis with the SMALLEST possible change:

1. Make ONE change
2. Run the reproduction steps
3. Did it fix the issue?
   • Yes → Proceed to Phase 5
   • No → Return to Phase 2 with new information. Do NOT add more fixes.

Critical: If 3 hypotheses fail, STOP. The problem is likely architectural, not a simple bug. Discuss with the user before attempting fix #4.

Phase 5: Fix

Now — and only now — implement the proper fix:

1. Write a failing test that reproduces the exact bug
2. Implement the fix — address the root cause, not the symptom
3. Verify the test passes
4. Run the full test suite — ensure no regressions
5. Save the learning — record what you found for future reference:

   node ${CLAUDE_PLUGIN_ROOT}/tools/learnings-manager.mjs save --title "Root cause of webhook 500" --body "The webhook handler wasn't awaiting the database write, causing a race condition with the response" --tags "debugging,webhooks,async"
   

Red Flags

If you catch yourself doing any of these, STOP and return to Phase 2:

• "Let me just try this quick fix"
• "It's probably X, let me change it"
• "I'll add multiple changes and see which works"
• "I don't fully understand but this might work"
• "Here are 3 possible fixes" (without investigation)
• Proposing solutions before tracing the data flow

Integration with Guard

For critical systems, activate /guard before investigating to prevent accidental changes:

/guard → /investigate → fix → /canary

Guard ensures no destructive commands run during investigation, and canary verifies the fix in production.

Relationship to Systematic Debugging

This skill shares principles with ultraship:systematic-debugging but adds:

• Module freeze — scoped investigation prevents sprawl
• Learning capture — every investigation produces a learning
• Guard integration — safety during critical system debugging
• Escalation protocol — clear rules for when to stop and rethink