Authorization policy does not seem to apply to the Evaluation API

[lawmatsuyama]

Hi!

I'm trying to restrict access to namespaces using a custom policy.rego in Flipt v2.

My use case is to have static tokens scoped to specific namespaces. For example, I have a static token with the following metadata:

metadata:
  mode: readonly
  namespace: product

And my authorization policy looks roughly like this:

package flipt.authz.v2

import rego.v1

default allow := false

token_mode := input.authentication.metadata.mode
token_namespace := input.authentication.metadata.namespace

allow if {
    token_mode == "readonly"
    input.request.action == "read"
    input.request.namespace == token_namespace
}

allow if {
    token_mode != "readonly"
}

For Management API requests, I would expect this policy to prevent the token from accessing namespaces other than product.
However, when using the Evaluation API:

POST /evaluate/v1/boolean

I can evaluate a flag from the default namespace using the token whose metadata contains:
namespace = product

The request is successfully authenticated, but it looks like the authorization policy is not evaluated for this endpoint.

My expected behavior would be:

token namespace = product

/evaluate/v1/boolean with namespaceKey=product
→ allowed

/evaluate/v1/boolean with namespaceKey=default
→ forbidden

Is this expected behavior in Flipt v2?

[github-actions]

Issue Health Check: Well scoped

Type: Bug report

Well-scoped bug report that authorization policies are not enforced on the Evaluation API (POST /evaluate/v1/boolean) in Flipt v2. The issue includes the custom Rego policy, token metadata configuration, expected vs actual behavior, and a clear reproduction path.

Thanks — this issue looks well scoped and ready for the maintainers to evaluate.

Need help with Flipt? Support the project through GitHub Sponsors or learn about Flipt Pro.

[dosubot]

Hi @lawmatsuyama! I'm Dosu and I'm helping the Flipt team.

This is expected behavior — the Evaluation API intentionally skips authorization. The evaluation server implements SkipsAuthorization returning true [1], which causes the authorization middleware to bypass OPA policy checks entirely for evaluation endpoints.

The authorization docs reference policies applying to /api/v1/ requests (the Management API) [2], and evaluation endpoints are excluded by design.

For your use case of restricting evaluation to a specific namespace, the intended mechanism is namespace-scoped authentication rather than OPA policies. The evaluation server's AllowsNamespaceScopedAuthentication method returns true, which enables Flipt's NamespaceMatchingInterceptor to enforce that a static token can only access the namespace specified in its metadata. This interceptor runs at the authentication layer, so it applies to evaluation requests.

So your static token with namespace: product metadata should already be restricted to the product namespace for evaluation calls — if you're seeing it succeed for namespaceKey=default, that would be worth investigating further. Could you confirm how your static token is configured (specifically whether the namespace field is set under authentication.methods.token.bootstrap in your Flipt config)? The namespace scoping for static tokens relies on specific configuration to activate the NamespaceMatchingInterceptor.

If you specifically want OPA policy evaluation to also cover the Evaluation API, that would be a feature request — the current architecture deliberately separates the two.

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.