Show HN: A control-theory approach to detecting LLM agent instability

Category: library

Tags: llm-agents, control-theory, stability-monitoring

Score: 7.5/10 (Innovation: 7, Technical: 8, Documentation: 8, Utility: 7)

State-harness applies control theory (Lyapunov stability) to detect and classify failure patterns in multi-turn LLM agents, such as token spirals and retry storms, without additional LLM calls. It provides actionable diagnostics and integrates with popular frameworks like LangGraph and CrewAI, making it a novel tool for improving agent reliability and observability.

Target audience: backend devs, ai-engineers, devops

Repository: https://github.com/vishal-dehurdle/state-harness · Python · NOASSERTION · 16 stars

View on Hacker News