Definition
A safety layer that watches for an AI agent looping or going nowhere and forces it to rethink.
In Life-Harness, the runtime layer monitoring recent action history for repetition or lack of progress and injecting recovery prompts when trajectory degeneration is detected.