Operator identity
Required for any control change, and recorded as self-asserted and unverified — this platform has no login, so nothing here authenticates you.
Agent controls
Cost ceilings
No ceiling is configured, so nothing is enforced. Spend is still recorded and visible on the Costs page.
manage_budgets not granted
What these controls do and do not do
- Disabling an agent prevents its next invocation. The orchestrator checks before each agent starts, and a blocked agent is recorded as a skipped step with your reason — distinct from a cache reuse, which records a completed step because work really was reused.
- A cost ceiling is measured against accumulated recorded cost. It cannot prevent the action that crosses it: an action's cost is only knowable after the provider returns usage. No pre-execution estimate is used, because none exists.
- Publishing and notifications each carry a second, independent config flag. Enabling those agents here does not switch that flag on — both must permit an action.
- Every change is written to the append-only audit log with the capability used and your (self-asserted) identity.
Still not available
No user accounts or roles exist. Capabilities are granted to the shared API key through server configuration, so a grant applies to everyone holding that key — it cannot distinguish individual people. Notification preferences, scheduling controls and multi-site settings do not exist either. None of these are shown as empty forms, because an empty form implies a setting that can be saved.