The combined controls, approval gates, action budgets, monitoring, kill switches, structured stop reasons, and audit logging, that govern an AI agent's autonomous operation.