Gate Agent-proposed Actions Behind An Independent Authorization Boundary

TypeScript reference architecture enforcing an authorization boundary for AI agents: agents propose immutable intents, an independent policy engine returns ALLOW/ESCALATE/BLOCK, verified human approval issues a single-use execution grant, and a SafeExecutor runs only granted actions with verifiable audit dossiers.

AI Agents | TypeScript · agent-permissions · agent-safety