I applied Lyapunov stability theory to detect when LLM agents spiral

This project applies Lyapunov stability theory from control systems to detect when large language model (LLM) agents enter unstable, self-reinforcing error spirals during multi-step reasoning. Designed for AI safety researchers and developers of autonomous LLM systems, it offers a mathematically rigorous early-warning mechanism for cascading failures. It is interesting because it bridges classical dynamical systems theory with modern AI, providing a principled, real-time diagnostic tool for agent reliability beyond statistical anomaly detection.