control-theory

State-harness detects LLM agent instability using control theory

Open-source state-harness library uses a Lyapunov energy function to catch token spirals in multi-turn LLM agents and explain failures without extra LLM calls.