--:--:-- UTC

Policy

Approval automation, platform enforcement, and team guardrails.

Approval rules

Platform policies

Policy	Outcome	Details
Granted tool access	Allows a tool call when the fleet has the connector and scope granted, and the organization has enabled the connector.
Blocked team actions	Blocks a tool call when the team blocks that scope or all scopes on the connector.
Budget enforcement	Blocks actions with side effects when the remaining budget is exhausted or below the estimated cost.
Verified caller identity	Blocks calls without a custody chain, with more than eight custody links, or with an unsupported root identity. Roots must be a user, API key, or service account.
Read-only actions	Allows read-only (PURE) calls without human approval.
Reversible actions	Allows REVERSIBLE calls without human approval unless the team requires stricter approval for that class.
Human approval for irreversible actions	Always blocks unattended IRREVERSIBLE calls. No rule can override this policy.
Fleet approval requirements	Requires human approval when the fleet manifest marks the scope as requiring approval.
Stricter team approvals	Provides no default unattended permission for classes marked as stricter by the team.

Signed policy bundle

Signed with Ed25519. Bundles refresh every 30 seconds. Effectful calls fail closed after 24 hours without a valid bundle; PURE calls remain allowed.

Team guardrails

Effect classes

PURE: Read-only actions. No approval by default.
REVERSIBLE: Actions that can be undone. Allowed by default unless the manifest or team requires approval.
COMPENSATABLE: Actions requiring a compensating operation to recover. Human approval by default; a saved rule can allow them.
IRREVERSIBLE: Actions that cannot be undone. Always require a human; rules cannot auto-approve them.