Remote approvals, policy checks, and execution evidence for unattended AI agents.