Investigate

1. Capture the symptom.

   • Record the exact error, failing test, or unexpected behavior.
   • Note reproduction steps and the smallest affected input.
   • Check recent history with git log --oneline -10 or a narrower range.

2. Form three hypotheses.

   • Write exactly three plausible root-cause hypotheses.
   • Make each hypothesis specific enough to falsify.
   • Avoid vague claims like "the code is wrong".

3. Verify each hypothesis.

   • Reproduce the failure with a focused test or script.
   • Trace the relevant code path.
   • Compare expected vs actual behavior.
   • Record evidence that confirms or rejects each hypothesis.

4. Report the result.

   • Summarize the symptom, hypothesis results, and the most likely root cause.
   • List the fix direction and the regression test that should be added.
   • If no fix was applied, say that explicitly.

Working rules

• Treat src/, lib/, and other production directories as read-only during investigation.
• Use tests/debug/ or another temporary test location for repros when helpful.
• Prefer small, evidence-driven changes over broad rewrites.
• If multiple agents are available, assign one hypothesis per agent and merge the evidence.

Coordinator follow-up

• Use pipeline-coordinator only after evidence has narrowed the problem and multiple fix strategies remain.
• Do not replace root-cause investigation with coordinator speculation.

Output shape

Produce a concise investigation report with:

• symptom
• reproduction
• three hypotheses
• evidence for each hypothesis
• root cause
• fix recommendation
• regression test recommendation