Lists unhandled-escalation events across the tenant, newest first.
Defaults to the last Fruxon.Model.HumanTasks.UnhandledEscalationReadService.DefaultLookback (these events are terminal and never drain, so an unbounded all-time list would stop being an attention signal); pass `sinceMs` to widen or narrow the window. Each item carries its `agentId`, `topicId`/`participantId` (for deep-linking into the conversation), and the failure reason. Read-only — the runtime already degraded the escalation gracefully.
Authorization
Bearer JWT Authorization header using the Bearer scheme. Enter 'Bearer' [space] and then your token.
In: header
Path Parameters
The tenant identifier
Query Parameters
Optional lower bound (Unix ms) on the event time. Defaults to now minus the lookback window.
int64The maximum number of items to return per page
int32int32Opaque cursor token from an external system (e.g. RAG API). When set, M:Fruxon.Common.Collections.PageToken.AsString returns this value directly instead of encoding Fruxon.Common.Collections.PageToken.Skip.
Response Body
curl -X GET "https://api.fruxon.com/v1/tenants/string/unhandledEscalations"{
"items": [
{
"id": "00000000-0000-0000-0000-000000000000",
"agentId": "string",
"networkId": "00000000-0000-0000-0000-000000000000",
"sessionId": "string",
"topicId": "00000000-0000-0000-0000-000000000000",
"participantId": "00000000-0000-0000-0000-000000000000",
"detail": "string",
"createdAt": 0
}
],
"nextPageToken": "string",
"totalCount": 0
}{
"type": "string",
"title": "string",
"status": 0,
"detail": "string",
"instance": "string",
"property1": null,
"property2": null
}Records the human decision and reopens the parent task for worker resume
On approve, the parent execution is requeued from the parked step and the tool call proceeds with its original arguments; on reject, the rejection text is fed back into the flow as the tool result so the agent can react to (or surface) the operator's reasoning. The responder identity is captured from the caller's auth context, not the request body, so audit trails reflect the real user. Requires Admin role on the agent to prevent arbitrary viewers from authorizing privileged tool calls; to abandon a request without resuming, use `POST .../pendingApprovals/{pendingApproval}:cancel`.
Lists all available alert types with labels and descriptions.
Returns the static catalog of alert types an agent can fire — these are compiled into the server via `AlertTypeRegistry`, not tenant-scoped or persisted. Use the returned identifiers when classifying alerts on agent runs so the UI can map them to friendly labels and severities. The list is stable across requests; cache aggressively client-side.