Show HN: Find where multi-agent AI systems break before production (github.com)

A new tool called "swarm-test" has been launched to help developers identify potential failures in multi-agent AI systems before they go live. This tool performs static reliability testing on frameworks such as CrewAI, LangGraph, and AutoGen—offering a cost-effective alternative by eliminating live API calls. It quantitatively assesses the reliability of agent networks, revealing vulnerabilities like silent cascade failures and single points of failure (SPOFs). By analyzing the topology of connected agents, swarm-test generates insights into system health, offering a "Swarm Score" and interactive visual representations of the agent architecture. The significance of swarm-test lies in its ability to promote more robust AI systems by proactively identifying weaknesses that could lead to serious failures in production environments. Key features include the detection of context leakage, intent drift, and timeout resilience, while also enabling users to generate reports that show historical trends in system performance. With a customizable configuration and easy integration into continuous integration (CI) workflows, swarm-test not only enhances reliability but also facilitates improved collaboration among developers in the AI/ML community. The tool’s open-source nature allows for community contributions, further driving innovation in AI system testing.