FailureDNA: Agent memory that knows when not to trust itself

A developer built FailureDNA, a persistent memory system that tracks real outcomes of past incident responses. It uses a deterministic validity gate to classify past actions as 'use', 'inspect', or 'avoid' before an LLM can act. The system prevents agents from repeating known failures by removing invalid actions from the candidate list.