Make Your AI Agent Better, and Prove It
PREMIUM. The loop I use to make AI agents, prompts and skills better, and prove it: act, get corrected, write the rule, re-test, promote or retire. One prompt builds a test set from your own past outputs, measures a baseline, makes one change and judges it blind. You get a round report that says kept or reverted. 11 lessons with my real file layouts, judge brief, thresholds and the rounds that lost. Two Claude Code skills included.