Request a fresh AI triage submission for a host.

The request is queued ahead of automatic work and ignores the configured minimum interval between verdicts, the unchanged-input check and the reuse of another host's verdict, so for a host with something alerting it always results in a submission to the configured model. A host with nothing alerting is recorded as clean without a submission, as it would be automatically. The configured daily limit does not apply to the request, though its submission is counted against the day. A request for a host whose verdict is being formed at that moment runs again once it finishes, against the host as it is then; a scan that lands during a requested run re-runs the host as automatic work rather than as a second request.

The host tag exclusion does not apply to a request. exclude_host_tags scopes automatic triage, so a host it keeps out of the unattended path is still triaged when an operator asks, and the verdict that comes back is kept rather than taken back by the next automatic pass. While the request waits, the host's existing verdict, if any, stays in place and is reported with the request's queue position beside it.