skill-optimizer

Skill optimization against benchmarks uses validation-gated mutations applied to the execution body, evaluating improvements through robust median-of-3 checks and epsilon gating to ensure reliable performance gains.

Validation-gated mutation for skill tuning applies edits to the execution body only after passing robust median-of-3 benchmark checks with epsilon gating, preventing regressions and ensuring reliable improvements.

Bootstrapping a new evaluation suite involves generating starter benchmarks from routing data or skill definitions, then splitting the data to create robust held-out test sets for evaluating machine learning skills.

Yes, you can optimize a skill without losing routing behavior because the optimization process preserves the routing surface by exclusively applying gated mutations to the execution body.

Skill mutations are applied atomically with versioning and audit trails to prevent regressions, ensuring every optimization change is tracked and validated against held-out benchmark sets.

Use epsilon gating during skill benchmarking when you need to verify that mutations provide a statistically meaningful performance improvement before committing the changes to the execution body.