Your Agent Eval Set Is Rotting: Build a Failure-Mining Loop for Google ADK

An agent evaluation set starts healthy. It contains the obvious intents, a few tool failures, and the...