Tag: fundamentals

An agent eval is a repeatable test that combines an input, expected behavior, execution evidence, and a grader.

A trace records how an agent reached an answer, including model calls, tool use, handoffs, retries, and verification.