📄️ Agent assurance
Requires the agent behind a call to be verified (the gateway forwarded the agent's own token) rather than merely asserted by the gateway.
📄️ Agent owner required
Refuses a decision about an agent that has no named human owner on its registration.
📄️ Amount limit
Refuses a call whose stated amount is over a ceiling. When several layers set a ceiling, the smallest one wins.
📄️ Blocked intents
Denies a RAR entry whose intent type an operator has listed as blocked.
📄️ Blocked scopes
Refuses the listed scopes for the calls its layer holds for, whatever the client was granted. Its first use is refusing agents that carry no registered kind.
📄️ Sensitive capability opt-in
Denies operations a connector has named as sensitive — creating a public sharing link, inviting someone outside the tenant, posting into an external chat — until an operator enables that capability for that resource.
📄️ Human approval (CIBA)
Requires human approval for a RAR intent matching an operator rule, by emitting an approval_required subject obligation that kicks the caller into the CIBA flow.
📄️ Client entitlements
Resolves what a client may do — its grant types and granted resources — from the highest-precedence identity source that matches it.
📄️ Connector-authored policy
Composes deny rules authored by installed connectors into the platform decision, normalising their shape.
📄️ Consent PIP check
Asks an external consent PIP whether the subject has consented to this client, and builds the consent obligation that sends them to the consent IdP when they have not.
📄️ Denylist
Refuses a client, subject, redirect domain, software id or issuer that the operator has listed, at every entrypoint that can see it.
📄️ DPoP proof requirement
Requires a valid DPoP proof at /token, for clients that registered for sender-constrained tokens and for any client the operator names.
📄️ Watch-only mode
Lets an operator put a resource type into watch-only mode: the decision API records what policy would have denied and lets the call through.
📄️ File access control
Decides which files, documents, spaces or calendars a connector may return or mutate, by label, by container location, or by the resource's own id.
📄️ Gateway obligation enforcement
Carries out the obligations the PDP emits, and fails closed when it cannot.
📄️ Connector IdP routing
Sends a subject to the IdP a managed resource forwards credentials from when the AS holds no forwardable token from it, and keeps that resource out of a token issued without one.
📄️ IdP-authored policy
Composes the rules and scope config an IdP ships in its policy bundle into the platform decision and into which claims a leg may assert or a client may see.
📄️ IdP routing
Chooses which identity provider an authorization request is sent to.
📄️ JIT single-use tokens
Makes an issued token single-use for SPIFFE workloads: the AS revokes it at /introspect after the first successful check.
📄️ PKCE requirement
Requires a PKCE code challenge at /authorize — for every client, for public clients only, or for a named list.
📄️ Public client policy
Bounds what a dynamically-registered or CIMD client may ask for at REGISTRATION: which grant types, scopes, resource indicators and client authentication methods are permitted, and whether open registration is allowed at all. It does not decide what an already-registered client is entitled to — that is client_entitlements, scoped by a binding.
📄️ Record scope
For listed scopes, the record being touched must be on a list the caller passes with the call, so an agent can only reach the records it is serving.
📄️ Operator read-only mode
Lets an operator put any gateway-fronted resource into read-only mode, denying every scope that mutates upstream state regardless of what the client was granted.
📄️ Resource type registry
Admits only resource and RAR intent types the platform knows, so that every admitted type can have policy hooked off it.
📄️ Software statement validation
Requires a software statement at registration when the operator demands one, and requires any statement presented to declare its grant types and resources explicitly.
📄️ SPIFFE workload identity
Admits SPIFFE-identified workloads from trusted domains only, and grants each one entitlements by its SPIFFE path.
📄️ Subject assurance
Requires the user behind a call to be verified (the caller passed the user's own IdP token) rather than merely asserted by the gateway.
📄️ Subject restrictions
Denies a named subject specific resource types or actions.
📄️ Token derivation limits
A token derived from another token — by introspection re-issue or by token exchange — may never carry more than the token it came from.
📄️ Token TTL override
Shortens the lifetime of an issued token for SPIFFE workloads. Composes as a minimum across every contributor, so it lowers a TTL and never raises one.
📄️ Trusted issuers
Rejects a software statement or subject token whose issuer is not on the operator's trusted list.
📄️ URI validation
Rejects a redirect or client URI that does not parse.