| name | debug |
| description | Debug a failing runtime, test, or integration path by collecting concrete evidence before proposing fixes. |
| when_to_use | Use when behavior is failing, flaky, slow, or unclear and you need to diagnose the cause from logs, traces, config, or reproducible commands. |
| user-invocable | true |
| disable-model-invocation | true |
Debug
Do not guess. Build a minimal evidence trail.
Goal
Find the concrete failure mode and the most likely root cause.
Workflow
- Reproduce the issue with the smallest command or scenario available.
- Collect direct evidence:
- failing command output
- relevant logs
- config values involved in the path
- recent code paths touched by the failure
- Narrow the scope:
- identify whether the problem is input, config, environment, dependency, or code logic
- isolate the first bad step, not just the final symptom
- Only then propose or implement a fix.
Output requirements
- reproduction step
- evidence collected
- most likely cause
- next fix or mitigation
Guardrails
- Prefer focused log reads and targeted searches over broad repo-wide speculation.
- If the issue is intermittent, state what is confirmed vs inferred.
- If you cannot reproduce it, say that explicitly and list the missing evidence.