skill-tuning

Closed-loop, low-cost skill-tuning pipeline: a cheap model mass-generates adversarial evals; a capable model surgically tunes; curated scenarios become a public eval set.

Tunes triggers and boundaries with Tune-Skill, then polishes discovery with Findable-Polish — use it only when you want both halves at once. A triggering-only fix belongs to tune-skill, a listing-only one to findable-polish. Not for creating skills (use draft-skill).