arielshad/balagan-agent

Applies controlled fault injection to AI agent tool calls across multiple failure modes—including timeouts, malformed responses, hallucinations, context corruption, and budget exhaustion—while measuring recovery metrics like MTTR and reliability scores compatible with SRE standards. Ships with built-in integrations for CrewAI and a CLI for scenario-based testing, plus a modular injector architecture supporting tool failures, latency patterns, and data corruption. Provides rich observability through terminal, JSON, Markdown, and HTML reporting to help teams validate agent resilience before production deployment.

Available on PyPI.