Learning Loop (Compounding Agent Work)

Use when agents keep making the same kinds of mistakes across PRs, when review comments on agent work repeat, when many agent PRs are rejected or abandoned, or when a team asks how to make each unit of agent work improve the next one.