Optimizing

It runs the audit suite, then systematically tightens descriptions for better triggering, cuts token bloat, and fixes structural issues like content in the wrong layer or workflows that should be separate skills. The execution flow is rigid (diagnose, decide, delegate to authoring, verify), but target selection adapts to what the audits surface. Notable feature: includes an A/B eval protocol for description changes so you can measure whether your "improvement" actually regresses trigger accuracy. Most useful after initial development when you're polishing for production or responding to user feedback that a skill isn't firing when it should.