Action: Bughunt

Available in: Full mode only

Read shared.md for resume state schema before starting. Check for existing state in .claude/pm-state.json.

Step 1: Identify the Bug

Option A -- GitHub issue number provided (e.g. #55): Fetch with gh issue view. Save state. Proceed.

Option B -- No argument: List open bugs: gh issue list --label bug --state open --sort created Let user pick one, or describe a new bug.

Option C -- Description provided (no GitHub issue): Offer to create GitHub issue for tracking, or investigate without.

Step 2: Gather Evidence

Ask progressively (not all at once):

1. Error output / stack trace?
2. Relevant logs?
3. Steps to reproduce? Intermittent?
4. Observability data (metrics, traces, alerts)?
5. Environment?
6. Recent changes / deploys?

Save state. Offer: "Evidence collected. Ready to investigate, or save & exit?"

Step 3: Investigate

1. Reproduce locally (run tests, dev server)
2. Trace the code path through the codebase
3. Form hypotheses ranked by likelihood
4. Test each hypothesis:
   • Describe what you're checking and why
   • Rule in or out with clear reasoning
   • If using GitHub, add comment after each hypothesis
   • Save state after each. Offer: "Hypothesis [N] tested ([status]). Continue, or save & exit?"
5. Iterate if all ruled out -- ask for more context, broaden search

Step 4: Root Cause Identified

1. Present finding: root cause, affected code, fix approach

2. Save state. Ask user:

   • Fix it now -- TDD: create branch, write failing test, implement fix
   • Just document it -- update GitHub issue with root cause
   • Save & exit -- save findings, resume later

3. If fixing (TDD):

   • Create branch, write failing test, commit. Save state.
   • Implement fix, commit. Save state.
   • Run full test suite
   • Add summary comment to GitHub issue

Step 5: Wrap Up

1. Summarize findings
2. Update Notion board card -- set Status to Done (or In Progress if PR pending) via notion-update-page
3. Offer to create issues for related bugs found
4. Delete state file

Key Rules

• Developer drives -- present findings, wait for decisions
• Evidence-based -- every hypothesis cites specific evidence
• Rule things out explicitly -- document what was checked
• TDD for fixes -- failing test first, no exceptions
• Save state often -- after every checkpoint
• GitHub is optional for tracking -- never force it