Test and improve Agent Skills with real your AI Code sessions: scaffold eval suites, measure what a skill adds, catch trigger collisions, and keep only fixes that pass a regression gate
3 in the Claude listing