nucleated status
The view Watch would show, read on the box itself. No account, no network, nothing sent anywhere.
nucleatedOn this page
Run it #
nucleated status
agentbox · disk 41.5% · load 0.62 · up 14d 3h · memory pressure 3.1%
Watch: reporting to https://api.nucarium.com for Your Org · last heartbeat 22s ago
Daemon: running
Needs you
client-a 7f3c2a1e permission 6h 4m your-org/web, your-org/api workspace/fix-login
client-b 1b9d04c7 input 22m your-org/api workspace/bump-deps
Failed
client-a c41e88a2 rate_limit 1h 12m your-org/web workspace/nightly-tidy
Down
client-c agent server is not running
Working
client-b 93fa6d10 9m your-org/web, your-org/api workspace/add-search
Idle
client-a 5e2b7c39 31s your-org/web workspace/docs-pass
Environments
client-a agent server running /home/agent/workspace
client-b agent server running /home/agent/workspace
client-c agent server not running /home/agent/workspace
Housekeeping
refresh 14m ago 2 fetched
cleanup 3d ago 3 kept, 1 removed
snapshot 14m ago no worktrees
The groups are in the order the inbox sorts them: what needs you first and longest first, then failed, then anything down, then the rest. A group with nothing in it is left out.
The command reads the daemon's own record of the box. It does not ask Watch anything, and it works the same on a box that has never signed in.
What each part says #
| Line | Where it comes from |
|---|---|
| The first line | The machine's name and the same numbers the heartbeat carries. Memory pressure is shown where the box can measure it. |
Watch: |
Whether the box is reporting, to which account, and when Watch last accepted a heartbeat. Anything waiting in the spool, or the last error from Watch, is added to the end. |
Daemon: |
Whether the daemon is running. If it is not, the sessions below are as it last saw them. |
| A session | Its environment, the first eight characters of its session id, what it is waiting for or how it failed, how long it has been that way, its repos, and its branch. |
Down |
An environment whose agent server is not running. |
Environments |
Every environment in the config, whether its agent server is running, and its nucleus. |
Housekeeping |
When each job last ran and what it did. |
The Watch: line is the first place to look when Watch and the box disagree. It reads one of:
| It says | Meaning |
|---|---|
reporting to … · last heartbeat 22s ago |
Signed in and getting through. |
reporting to … · no heartbeat accepted yet |
Signed in, and nothing has got through. |
not signed in. Nothing leaves this box. |
No token. Run nucleate login to connect it. |
signed in, but target is "none". Nothing leaves this box. |
The config is keeping it quiet. See Targets. |
the token is no longer accepted. Run nucleate login. |
Watch rejected the token: it was revoked or replaced. |
the token file is readable by others, so it is not used. Make it 600. |
Fix the file's permissions. |
What it cannot tell you #
Watch works out some things from the outside that a box cannot know about itself:
- Whether the box is stale or down. Those mean "Watch has not heard from it", and a box that can run this command has not gone dark.
- Whether an alert has fired.
- Anything about your other machines.
So nucleated status is the whole picture for one box, and Watch is the picture across all of them, including the ones that have stopped talking.
As JSON #
nucleated status --json
{
"daemon": {
"pid": 63667,
"running": true,
"updated_at": "2026-11-19T02:40:56.009Z",
"version": "0.3.0"
},
"environments": [
{
"agent": "claude-code",
"name": "client-a",
"root": "/home/agent/workspace",
"server_alive": true,
"sessions": { "needs_input": 1 }
}
],
"generated_at": "2026-11-19T02:41:07.955Z",
"housekeeping": {},
"inbox": [
{
"attention": "needs_input",
"branch": "workspace/fix-login",
"environment": "client-a",
"error_kind": null,
"kind": "session",
"last_event_at": "2026-11-18T20:37:03.980Z",
"needs_kind": "permission",
"repos": ["your-org/web", "your-org/api"],
"session_id": "7f3c2a1e-58b4-4d0c-9a6e-2b1f0c9d8e7a",
"since": "2026-11-18T20:37:03.980Z",
"started_at": "2026-11-18T20:21:55.951Z",
"status": "needs_input",
"workspace": "fix-login"
}
],
"machine": {
"disk_used_pct": 41.5,
"load_1m": 0.62,
"memory_pressure_pct": 3.1,
"name": "agentbox",
"uptime_s": 1220400
},
"now_ms": 1795056067955,
"watch": {
"account_name": "Your Org",
"api_url": "https://api.nucarium.com",
"events_error": null,
"heartbeat_error": null,
"last_heartbeat_at": "2026-11-19T02:40:45.000Z",
"pending_events": 0,
"rejected_events": 0,
"signed_in": true,
"target": "hosted",
"token": "valid"
}
}
inbox is the same list the text view prints, in the same order, with one entry per session and one per environment that is down. The field names match the ones Watch's API uses for the same things. workspace is the session folder's name: it is here because this is the box talking to itself, and it is not something Watch receives.
watch.pending_events is how many notes are waiting in the spool, and rejected_events how many Watch has refused for good.
Keep it on screen #
nucleated status --watch
Redraws every five seconds until you press Ctrl+C.
From your workstation #
It is an ordinary command, so it runs over SSH:
ssh agentbox-agent nucleated status