What problem does it solve? Agent skills accumulate stale content, weak triggers, and rules that never fire, and reading a skill's text cannot reveal whether it actually runs in real sessions. This Skill turns lessons from real agent sessions into concrete, tested improvements for personal skills stored in ~/.agents/skills. ## Core Features & Use Cases - Firing Audit: Runs audit.sh against session transcripts to report per-skill load counts, never-fired skills, gate compliance rates, and where in each session a skill fires. - Blinded Comparison: Defines a protocol for A/B testing skill edits with blind variants and a single judging pass before promoting a change. - Editing Discipline: Enforces rules like the no-op test, description-vs-body labeling, sediment removal, and harness-agnostic wording when drafting skill changes. - Use Case: After noticing an agent repeatedly skips a verification step, run the firing audit to confirm the relevant skill never loads, then fix its description trigger rather than adding more body prose. ## Quick Start Ask the agent to audit which of your personal skills actually fired in recent sessions and propose improvements based on the results.
