| name | diagnose-pro |
| description | Disciplined debugging and diagnosis workflow for hard bugs, regressions, flaky behavior, performance issues, and unknown failures. Adapted from Matt Pocock-style diagnosis loops for Codex v4.1. Use when root cause is unclear or the bug is not fixed by a direct local change. |
Diagnose Pro
Workflow
- Reproduce or capture the failure as concretely as possible.
- Reduce the problem to the smallest failing path.
- Form one hypothesis at a time.
- Add observation: logs, assertions, tests, traces, screenshots, query plans, or minimal probes.
- Fix the root cause, not the symptom.
- Add a regression test or verification step.
- Remove temporary probes before finishing unless they are useful permanent diagnostics.
Diagnosis Rules
- Prefer evidence over intuition.
- Change only one variable per diagnostic step when possible.
- Treat "works on my machine" as an environment-difference bug, not a conclusion.
- For flaky bugs, collect timing, concurrency, cache, external API, and environment signals.
- For performance bugs, establish a baseline metric before optimizing.
- For security bugs, preserve exploit conditions and prove the boundary is closed.
Output
Use this shape:
根因：...
证据：...
修复：...
验证：...
If root cause is still unknown, state the strongest remaining hypothesis and the next concrete observation needed.
Boundaries
- Do not claim files, generated media, deployments, scans, account changes, or external writes unless the active tool or command output proves them.
- Do not override system, developer, user, project, or selected-skill rules.
- If another skill is more specific for the user's intent, hand off or use it as the main skill instead of duplicating its workflow.
