PoC-stage external execution gate.
AGENT APPROVAL / PUBLIC EXPLANATION
How to require approval before an AI agent acts externally
For automated emails, live deployments and orders, place an approval gate immediately before external effect. Do not allow an action to proceed without its stated approval conditions.
Approval means hold before proceeding
The proposal does not ban all autonomous work. It places a review boundary before operations affecting external systems. Requests without adequate approval are held or escalated rather than silently executed.
- Design aim: make unauthorized external actions stoppable.
Permit Token and reason codes
The source proposes issuing a Permit Token only when the request meets its approval conditions. Reason codes preserve why a request was stopped and allow later examination of why actions passed or did not pass.
- A described Permit mechanism is not proof that a live token was actually issued.
Make human review meaningful
The reviewer, review reason, authority and scope of risk need to be visible together. This links confirmation to responsibility rather than reducing it to an unexamined click.
- Human confirmation by itself does not guarantee safety.
Related sources and materials
Pre-execution architecture.
Japanese description of stop-reason recording.
A scoped proposal, not an operational guarantee.
Structure and evidence at the action boundary.
doc_id: C3-WEB-ANSWERS-APPROVAL-GATED-EN-0.1 / version: 0.1.0 / status: REVIEW_REQUIRED / translated: 2026-10-13 / source_doc_id: C3-WEB-ANSWERS-APPROVAL-GATED-0.1 / source_last_updated: 2026-06-05