Skills that keep learning your repo.

Every merged PR and failing test becomes a repo-specific fact, recalled into skills at load time — so your skills keep evolving and the agents using them get better with every change. No LLM in the learning loop. The skill files are never rewritten.