machin-watch

name	machin-watch
description	Run and configure machin-watch — a tiny uptime monitor (written in machin/MFL) that polls URLs on a per-watch interval and alerts on up<->down transitions via a machin-notify hub. Use this skill to monitor that a deployed service stays up, add/remove watches, run one-shot checks, or run the monitor daemon.

machin-watch — uptime monitor

Mental model

A daemon polls each watched URL on its own interval. A watch is up on 2xx/3xx, down on 4xx/5xx or a transport error (dns/connect/tls). State lives in SQLite (persists across restarts). On a transition (up→down or down→up) it POSTs an alert to a machin-notify hub — never on every poll. Built only on existing builtins (http_get, SQLite, sleep, raw sockets) — no new machin features.

Build

Needs the machin compiler, a C compiler, libsqlite3 (links OpenSSL for HTTPS).

./build.sh            # → ./machin-watch

Use

machin-watch add https://example.com --name site --every 60s --channel alerts
machin-watch list            # current up/down state per watch
machin-watch check           # one-shot check of all watches (prints + alerts on change)
machin-watch daemon          # run the loop (wakes every 5s, checks due watches)
machin-watch remove site

--every takes 30s/5m/1h/2d or a bare number (seconds). The first check of a brand-new watch alerts only if it's already down.

Alerts (via machin-notify)

Set the hub token; a per-watch --channel overrides WATCH_NOTIFY_CHANNEL.

WATCH_NOTIFY_TOKEN=<hub-token> WATCH_NOTIFY_CHANNEL=alerts machin-watch daemon

Messages: 🔴 DOWN — name (url)\nstatus 503 (or error: connect) and 🟢 UP — name (url) recovered after 2m. See machin-notify's SKILL.md for the hub.

Configuration (environment)

var	default	meaning
WATCH_DB	watch.db	SQLite watch list + state (use an absolute path when deployed)
WATCH_NOTIFY_TOKEN	—	machin-notify token (alerts disabled if unset)
WATCH_NOTIFY_CHANNEL	—	default channel; --channel per watch overrides
WATCH_NOTIFY_ADDR	127.0.0.1:48090	machin-notify daemon address

Deploy

systemd service, Restart=always, absolute WATCH_DB, and the notify env set — ExecStart=/path/machin-watch daemon.

Gotchas

• Watches both http:// and https:// (machin v0.39.0 added plain-HTTP to the client).
• Checks are sequential (each request ~5 s timeout); keep the watch list modest or expect cycle time to grow with the number of down/slow targets.
• Run the machin-notify hub first — machin-watch only alerts if it can reach the hub at WATCH_NOTIFY_ADDR with a valid token.