Diagnose provides a disciplined, repeatable workflow for finding and fixing difficult bugs and regressions. It teaches agents to build a fast, deterministic feedback loop (tests, HTTP scripts, or headless browsers), reproduce the failure, generate ranked falsifiable hypotheses, instrument to test those hypotheses, implement fixes with regression tests where possible, and perform cleanup and a post‑mortem. The emphasis is on measurable signals and automation so debugging scales.
Use this skill when a user reports a bug, says something is broken, describes incorrect output, or reports a performance regression. Also use it when reproduction is flaky — the skill explains how to raise reproduction rates and when to stop and ask for more artifacts (logs, HARs, recordings, or production access).
Well-suited for code-focused assistants and developer-facing agents (Claude Code, Copilot/Codex-style agents, GPT-based coding copilots) that can run scripts, tests, and debugging harnesses.
This skill has not been reviewed by our automated audit pipeline yet.