Last released Nov 7, 2026
A conformance benchmark for safe recovery from ambiguous tool outcomes in LLM agents