nodecode-gateway-guardian

Gateway Guardian: the gateway's health in one line — OK, DEGRADED or BLOCKED, with the fault counts behind it. A Nodecode rewrite of kawsthingz/kawsthingz-hermes-gateway-guardian by kawsthingz.

A chat bot can go quiet without an error: two pollers on one bot token (the messenger answers 409), a gateway that froze, a lock left behind by a restart, a delivery that never landed. The original watched one agent runtime's gateway log for exactly those faults and answered one state with the exact cause. This cell reads the same faults out of the log files you name, counts them, and answers the same three states.

Get it

Tell the agent "add the gateway-guardian cell", or run nodecode add nc://gateway-guardian.

Set it up

Nothing is required. With no gateway-guardian.logs named, the newest *.log under this home's logs/ and Logs/nodecode/ is read. Point it at a file when your gateway writes somewhere else:

"gateway-guardian": {
  "logs": ["~/.nodecode/logs/gateway.log"],
  "tail_lines": 400,
  "tail_bytes": 262144,
  "watch_processes": true
}

A systemd install keeps its log in the journal rather than a file: name a file, or redirect the journal to one.

Use it

gateway_guardian_status — the model's tool: one state (OK, DEGRADED, BLOCKED), the cause, and the count of each fault pattern (409/polling conflict, freeze, stale lock, missed delivery) in the lines it read. It names the files it scanned and the gateway pids it found; it never quotes a log line and keeps no message body.
gateway_guardian_recover — the approved checklist for status, restart-gateway, clear-stale-lock or rebind-telegram. It never restarts, kills or rebinds anything: it prints the steps the operator runs, and appends the chosen action to gateway-guardian/recovery-intents.log in this home.
/gateway-guardian — the same health in one line; /gateway-guardian recover restart-gateway prints that checklist.

From eval: (gateway-guardian:status) and (gateway-guardian:recover "clear-stale-lock"), both strings.

The states are the original's thresholds: three or more 409/polling conflicts read BLOCKED; any conflict, or two warnings of one kind, read DEGRADED; with watch_processes on, a clean log and no running gateway reads DEGRADED too (no gateway process found).

Run .github/test-cell.sh tools/nodecode-gateway-guardian from the cells repository.