Evaluates Databricks custom agents — `ResponsesAgent`, framework-wrapped (LangChain, LlamaIndex, AutoGen), and bespoke `pyfunc`. Verifies trace contract, runs `mlflow.genai.evaluate` on the deployed endpoint, classifies failures, and routes to lifecycle skills. Use when the surface is a custom agent and the failure class is not yet pinned. Do NOT use when the agent is Knowledge Assistant, Genie, Supervisor, or Document Intelligence — invoke that surface skill instead.
