Last released Sep 22, 2026
A stress-test benchmark that provokes MAST failure modes in LLM agent frameworks, and shows how much of the failure comes from the framework's own prompt scaffolding.