Define reusable agents — a goal, instructions, and an execution mode — and test them against the live orchestration pipeline. A test run returns the engineered prompt for your agent's task; for full planning outputs, use Production.