List the silent-failure findings the detectors reported across a workspace.
A finding says one run failed in a way its status does not show: a failure
mode (e.g. CF-01), the subject it concerns, a severity and a reason. These are
per-trace findings, NOT the `swallowed_error` alert rule kind (formerly
`silent_failure`) that the alert tools (list_alerts, list_alert_rules) manage.
Newest first, followed by a per-day count and a cursor when more match. The
trace_id column feeds get_agent_trace. For a CF-01 (tool_loop) finding the subject
is the tool name and the spans column lists the repeated calls in order; an info
one is likely polling (polling_as_info in update_alert_rule) and alerts nobody. For
a CF-03 (context_rot) finding the spans
column gives the earlier LLM call, the call that came back smaller and, when marked,
the compaction span; pass the first two to get_context_loss as from_span_id and
to_span_id to see what the context lost. When a detector reports nothing,
get_silent_failure_coverage says whether it has the telemetry to.
workspace: optional workspace id, name, or `org-name/workspace-name` when the
same name occurs in two of your orgs (defaults to your default).
failure_mode: optional catalog id, e.g. `CF-01`.
agent: optional exact agent name, as the agent column reports it.
state: `provisional`, `confirmed` or `retracted`. Omit for provisional and
confirmed; a retracted finding is no longer a failure and is listed only
when asked for.
since_hours: how far back to look, on when the failure was detected (default
168, at most 744).
limit: rows to return, at most 200 (default 50).
cursor: the cursor from the previous page; send the same filters with it.
Example:
> What failed silently this week?
```json
{ "name": "list_silent_failures", "arguments": {} }
```
> Confirmed CF-01 findings for checkout-agent in the last day.
```json
{
"name": "list_silent_failures",
"arguments": {
"failure_mode": "CF-01", "agent": "checkout-agent",
"state": "confirmed", "since_hours": 24
}
}
```