connector-action-authorization

Call connector_action_policy_status to review existing grants, summarize the proposed scope in plain language, then call grant_connector_action_policy with the connector, action kind, target account, and constraints like allowed_recipients and max_per_day.

Use revoke_connector_action_policy when the user asks to remove or narrow a grant. First inspect current policies with connector_action_policy_status so you target the correct existing grant.

No. Connection grants access, but action policies grant behavior. Sending, replying, labeling, spam triage, and deletion each require an explicit action policy grant before the agent can perform them.

Enable allow_autonomous only when the user explicitly asks for ongoing or autonomous behavior outside live chat, such as heartbeat-driven actions. Keep destructive actions like delete out of autonomous policy unless the user is exceptionally explicit with narrow constraints.

Prefer narrow constraints such as allowed_targets or allowed_recipients and max_per_day over broad grants. Add expiration when relevant, and use requires_per_action_approval false only when the user wants preauthorization instead of case-by-case approval.