KevinZai Operationalize Every Fix
When a bug is found and fixed, don't stop at the fix. Extract the full learning:
The Protocol
1. Test-First (Before Fixing)
Write a reproduction test FIRST:
Bug reported → Write failing test → Confirm test fails → Fix bug → Confirm test passes
The test proves the bug exists. The fix proves the test works. Don't reverse this.
2. Expand Test Coverage
After fixing, ask:
- What's the class of this bug? (off-by-one, null handling, async race, etc.)
- Write tests for the whole class, not just this instance
- Edge cases: empty input, max values, concurrent access, unicode, timezone
3. Sweep for Siblings
Check: does the same mistake exist elsewhere in the codebase?
# Example: if the bug was unchecked null access
grep -rn "\.property" src/ | grep -v "?." | grep -v "!= null"
Fix all siblings. Each one is a future bug prevented.
4. Update Instructions
If the bug reveals a gap in your workflow:
- CLAUDE.md — add a rule that would have caught it
- tasks/lessons.md — document the pattern
- Use corrective framing: "API responses must be null-checked before property access — verify this is happening in new code"
5. Root Cause Chain
Ask "why?" until you hit process:
Bug: user.name crashes on null user
→ Why? API returns null when user deleted
→ Why wasn't this handled? No null checks on API responses
→ Why? No rule about defensive API handling
→ Fix: Add rule to CLAUDE.md + tests for all API response handlers
Checklist (Post-Fix)
- Reproduction test written and passing
- Class-level tests added (not just this instance)
- Codebase swept for sibling instances
- Instructions updated if gap identified
- Root cause documented in tasks/lessons.md