| name | team-debug |
| description | Investigate a bug with competing hypotheses and parallel subagents. Use when the user asks to debug an error, find a root cause, or investigate why something is failing. |
Team Debug
Run a structured, read-first debugging pass before making fixes.
Workflow
- Capture the symptom, scope, reproduction hints, and any logs or errors the user provided.
- Read the relevant repo rules, the updated
.claude/agents/*.md ownership guidance, and references/hypothesis-testing.md.
- Generate 2 to 4 competing hypotheses across different failure categories when possible.
- Spawn read-only investigators from
.codex/agents/, choosing the smallest relevant subset of:
rust-backend-engineerreact-frontend-engineerdatabase-engineerdevops-engineerbackend-qafrontend-qadocument-writer for clarity, docs, or expectation mismatches when relevant
- Give each investigator one hypothesis, the files to inspect, and the confirming or falsifying evidence to look for.
- Require evidence with
file:line references and explicit verdicts such as confirmed, probable, inconclusive, or ruled out.
- Consolidate the results into:
- most likely root cause
- supporting evidence
- contradicting evidence
- recommended fix path
- remaining unknowns
Constraints
- The custom investigators are advisory-only and read-only. They inspect and explain; they do not edit files.
- Do not edit code unless the user explicitly asks to move from investigation to fixing.
- If the evidence is weak, say so plainly and explain what additional data would settle it.
