hanness-template

Audit Claude Code harness configuration (CLAUDE.md, settings.json, hooks, agents, skills, rules) against best practices. Use after adding/changing harness components, during periodic maintenance, or when you want to verify configuration quality. While entropy-detector checks code health, this skill inspects the harness configuration itself.

Record pitfalls and workarounds encountered during a session to docs/gotchas.md. Use after encountering errors or unexpected behavior, or before ending a session. Cleanup and staleness checks are handled by the entropy-detector agent.

Distill session learnings back into the repository before ending a session. Use at the end of sessions involving bug fixes, feature implementation, or debugging. Classifies learnings as gotcha (external pitfall) or harness (AI behavior improvement) and appends to appropriate harness components. Most effective when called just before commit/PR creation or as a completion signal.