Agents — /rbac/agents
The Agents page lists every actor that has any RBAC footprint in this workspace — every user, bot, integration, or delegated agent that holds a role assignment or carries a permission override. It is the "who has what" view, complementing the role-centric Assignments page.
Despite the name, this surface covers all four operational actor types (USER, BOT, INTEGRATION, DELEGATED_AGENT), not just AI agents. The fifth domain-model type, GROUP, is used internally for group-scoped assignments and does not appear here.
What you see
Stat cards
Four tiles summarize the current actor footprint (useAgents → actors query):
- Total actors — every actor that has at least one assignment or override
- Users — count of
USERactors - Bots — count of
BOTactors - Integrations — combined count of
INTEGRATION+DELEGATED_AGENT
Table
One row per actor:
| Column | Meaning |
|---|---|
| Actor ID | Truncated identifier (full value on hover) |
| Type | One of USER, BOT, INTEGRATION, DELEGATED_AGENT |
| Roles | Chips for the actor's role assignments (first three; +N more for the rest) |
| Overrides | Number of PermissionOverride rows attached directly to the actor |
| Effective permissions | Total count of resolved permissions, computed by the service |
| Actions | Eye icon → opens the per-actor detail (/agents/:actorId) |
Filters
- Search — case-insensitive substring match against the actor ID or type.
- Type filter —
All Typesor any of the actor types present in the result set.
Filters combine with AND.
Things you can do
| Task | How | Permission |
|---|---|---|
| Browse all actors | Scroll the list | canViewAgents (read access) |
| Search by ID or type | Use the toolbar | canViewAgents |
| Open an actor's RBAC profile | Tap the eye icon | canViewAgents |
| Revoke any of the actor's assignments | Use the Assignments page for the role in question | assignment:delete |
| Add or remove an override | Use the actor profile (actorPermissionOverrides + createPermissionOverride / removePermissionOverride) | override:write |
The list itself is read-only; mutations happen in the actor detail or via the role on Assignments.
Per-actor detail
Tapping the eye icon navigates to /rbac/agents/:actorId (the useRbacNavigate hook prepends the microfrontend's /rbac mount base to the /agents/:actorId argument from the list page). The detail surface renders the service's actorRbacSummary query and shows:
- The actor's role assignments (with scopes and grant timestamps)
- Direct permission overrides (
PermissionOverride) — bothGRANTandDENY, with optional expiry - The effective permission set (
effectivePermissions) — the same data the evaluator would resolve at runtime
If you want to know why an actor has access to something, the actor profile plus the Permission Checker are the two right tools — the profile shows what they have, the checker explains how it would resolve for a specific action.
Known UX gap. The eye icon currently lands on the agent list rather than a per-agent detail page — a small follow-up tracked separately. Use the Permission Checker with the actor ID in the meantime.
Tips
- The
Effective Permissionscolumn is the resolved count, not the sum ofRoles + Overrides. A deny override or a low-priority role can reduce the effective count below what the inputs suggest. - An actor disappears from the list when their last assignment and last override are removed — it does not represent membership in the workspace at large, only their RBAC footprint.
- For an aggregate "who runs the most checks" view, see Analytics → Top Actors.
- For the per-actor history of permission decisions (allow/deny over time), filter the Audit Log by actor ID.