Mirrors surfaces recurring errors from production AI agent runs and turns them into scoped corrections applied only where relevant. It previews which past conversations a correction would reach and measures the effect after activation.