Execution control layer for AI agents — prevents duplicate or incorrect real-world actions under retries, uncertainty, and stale context.
