I built this to test how production LangChain agents handle failures. It injects random exceptions into tool/model calls at configurable rates. Testing agent resilience is hard. This lets you simulate real-world failures (network issues, rate limits, service outages) in a controlled way.