What problem does it solve? Skills often fail to trigger because the router never sees their full description, the name uses camelCase or a cursed token, or the when-to-use text misses the phrasings users actually type. This Skill measures and fixes that routing surface so skills get invoked when they should. ## Core Features & Use Cases - Mechanical fixes: Converts descriptions to folded >- blocks, enforces word-separated hyphen-case names, and strips the claude token from names. - Eval-driven measurement: Generates should-trigger and should-not-trigger query sets, then measures invocation and false-positive rates with an isolated router agent. - Rewrite and verify loop: Rewrites name, description, and when-to-use for trigger coverage, re-measures, and reports before/after rates with a rename impact check. - Use Case: A skill fires only 10% of the time it should. Run the optimizer to apply mechanical fixes, generate evals, rewrite the frontmatter, and verify the invocation rate climbs toward the 80% target. ## Quick Start Optimize the skill at web-skills/my-skill so the router invokes it reliably, and report the before and after invocation rates.