The Alert Triage Desk (one lead instead of one session per alert)

What it does

Some alerts don't just sit in your inbox — Omniscio reacts to them by starting a Claude session to investigate and fix the problem. One stalled auto-lander can raise many different alerts at once, and each used to start its own session, so one fault became a crowd of near-identical sessions.

The triage desk changes that. While a desk crew has a live lead, every one of those alerts is handed to that one lead as a single queued message instead. The lead reads each one and decides:

• Same fault as something I'm already fixing? Fold it into that worker. Start nothing.
• A different fault? Start exactly one worker for it.

Nothing is merged automatically. The desk lead judges every alert itself, with the alert's title, its body, the brief the automatic fixer would have been given, and the list of related sessions already running in the same area.

When it is on

It is on by default and does nothing until a desk is live — a crew with a lead whose session is running. With no live lead, every alert behaves exactly as it did before, so an alert is never lost because the desk isn't there.

Two settings, both under Settings → Inbox:

• Send every automatic alert to the triage desk — the off switch. Off restores the old behaviour everywhere: one session per alert.
• Crew that handles alerts automatically — the desk crew's id. Leave it empty to use the crew that already sorts alerts; the id is the one shown on the crew in the Overseer panel.

The desk's identity is a crew, not a session, so replacing the lead — a hand-off to a new session that registers into the same crew — keeps the desk working with nothing to re-configure.

What the desk does with the card

When the desk takes an alert, the original card is archived and no "a session was started" receipt is posted, because nothing was paid for. The message itself carries the card's id, so the lead can still read the card. If the message could not be delivered — the lead is unreachable, in do-not- disturb, or the send failed — the card stays exactly where the old behaviour would have left it and the normal session starts instead.

For a script or a cron

GET /alert-triage-desk answers whether the desk is live, which crew it is, and which session leads it — so a scheduled job can ask before it starts anything:

curl -s http://127.0.0.1:19519/alert-triage-desk \
  -H "Authorization: Bearer $TOKEN" \
  -H "X-AMC-Source-Session-Id: $AMC_SESSION_ID"

{ "ok": true, "data": { "enabled": true, "crewId": "…", "crewName": "Alert Triage Desk",
                        "leadSessionId": "…", "live": true } }

live is the only field a caller needs to branch on.