Permission Checker — /rbac/checker
The Permission Checker is a "what-if" tool. Pick an actor, an action, and a resource, and the checker tells you whether the actor would be allowed — and why.
It is the fastest way to debug "why can't this user do X?".
What you see
A simple form with these fields:
| Field | What to enter |
|---|---|
| Actor | The actor you are testing — a user, bot, integration, or delegated agent. Pick from the dropdown or paste an ID. |
| Actor type | USER / BOT / INTEGRATION / DELEGATED_AGENT (the canonical enum values from the RBAC contract). The dropdown defaults based on what you picked above. |
| Resource type | The kind of thing being acted on (e.g. project, role, file). |
| Action | The verb (e.g. read, update, delete). |
| Resource ID (optional) | A specific resource. Leave blank to check at the workspace level. |
| Attributes (optional) | Free-form JSON for advanced checks (e.g. {"projectStatus": "archived"}). |
Then tap Run check.
What the result looks like
The result panel appears below the form:
- A green ✓ Allow or red ✗ Deny banner
- A breakdown of how the answer was reached:
- Roles considered — which roles the actor has and whether each one allows the action
- Overrides applied — explicit grants/denials at actor or role level
- Ownership — whether the actor owns the resource (and what that grants)
- Policy decisions — workspace policies that contributed to the result
- A timestamp and a copyable check ID for reporting
If the answer is Deny, the breakdown highlights the blocking rule so you know what to change to flip the result.
Bulk checker — /rbac/checker/bulk
A second view (/rbac/checker/bulk) lets you compose a list of actor / action / resource rows and get all the results at once. Use it for migrations or audits where you need to verify many cases.
The UI renders an editable grid (1–50 rows) with these columns per row: Actor ID, Actor type (User / Bot / Integration / Agent), Resource type, Action, Resource ID (optional). The + Add Row button is disabled at 50; the trash icon removes the row. Reset restores a single empty row.
Submitting runs checkPermissions once with all valid rows. The result list mirrors the input order and shows, per row: an allow/deny indicator, the resolved roleName (if one matched), an OVERRIDE_TYPE chip when a PermissionOverride was responsible (GRANT / DENY), and the decisionPath string for the matching evaluator step.
Per-request limit. The underlying
checkPermissionsmutation is capped at 50 inputs per request (RBAC_LIMITS.MAX_BULK_CHECK_SIZE; seemodules/rbac/specs/03-graphql-contract.md). The UI enforces the same cap by disabling Add Row at 50. For larger batches, run multiple submissions or script directly against the API.
Things you can do
| Task | How | Permission |
|---|---|---|
| Run a single check | Fill the form, tap Run check | read access to RBAC |
| Reset the form | Tap the Reset button | n/a |
| Copy a check ID | Click the ID in the result | n/a |
| Run many checks at once | Open /rbac/checker/bulk | read access to RBAC |
The check itself is read-only — it never grants or revokes anything. You can run it freely without affecting any actor.
Common use cases
- "Why can't user A delete project P?" — pick user A, action
delete, resource typeproject, resource IDP. The breakdown will show whether the user lacks the permission, has a deny override, or is blocked by an ownership rule. - "Will the new role I am about to publish actually allow what I think?" — assign the role to a test user first, then run the checker against that test user.
- "Did the migration land cleanly?" — paste your test cases into the bulk checker and verify the column of expected results matches.
Tips
- The check ID at the bottom of the result is sharable. If you escalate to support, paste the ID and they can re-run the exact same evaluation.
- The breakdown reads top-to-bottom in evaluation order. Stop reading at the first deny — that is what blocked the action.
- Attributes are case-sensitive JSON. If you mistype a key, the policy will silently not match. The result panel calls this out under "Policy decisions".
- For a deeper analysis, the Audit log under
/rbac/analytics/auditis the authoritative permission-decision stream — it records everycheckPermissioncall with its full decision path. The Activity tab on a role only tracks role-change events (creates / edits / assignments), not permission decisions.