Skill Forge

Drive a skill toward optimal through measured, iterative refinement — not a one-pass rubric check but a behavioral eval-and-improve loop. Use when building a new skill, hardening an existing one, deciding if a skill is good enough to ship, or proving a change actually improved a skill. Runs a two-tier eval (structural gate + behavioral outcome lift on held-out tasks, externally scored) inside a champion-promotion loop with non-gameable stops. Triggers include "make this skill optimal", "is this skill good enough", "harden/tune this skill", "did my change improve the skill", "evaluate this skill properly", or "build a skill to 85/100 and beyond".