Free agent · one file · MIT
Did your automation actually deliver?
The automation that fails loudly gets fixed the same day. The dangerous one keeps reporting success while it quietly stopped delivering. This is the watchdog that runs this company, about 90 scheduled jobs deep, pointed at yours.
What it asks
One question per automation, every hour: did the thing you built it for happen, on time? A file that must be fresh. A URL or JSON timestamp that must be fresh. An n8n workflow that must have succeeded recently, not just be "active". A log that must contain the success line and not the error line. Or any shell command you write.
What it does when the answer is no
Pages you on Telegram at once. Then again every 20 hours until it passes. Then one RECOVERED line. A pager that pages once is a noticer; this one nags. touch state/KILL silences it during planned work.
What you get on Monday
A delivery report: runs, failures, first failure time, and the list of automations that were silently dead the whole week. The ones that were probably still showing green somewhere.
Claude Code users
Open the folder and say "set up the watchdog". The /setup skill interviews you and writes the config; /watch explains a failure in one sentence; /report writes the weekly file.
Install, three commands
python3 watchdog.py setup # company, Telegram bot, the automations to watch
python3 watchdog.py check # run once, page on failures
python3 watchdog.py install # prints the cron line, writes a launchd plist on macOSSecrets never sit in the config: use env:VAR_NAME. Read every line of watchdog.py; it is short on purpose.
Why it exists
Every failure class it checks burned this fleet first. Five weeks of social distribution died behind a switch that paged once and then rotted. A customer waited sixteen days on a pipeline that logged "delivered". A vendor deleted an API mid-rebrand and the orchestrator logged every run as a success for weeks. That last one is written up as a case study.
If the weekly report turns up something that has been quietly dead for a while, that is the exact failure class we work on for clients: the Silent Failure Audit is $750 for a fixed one-week scope, and the reliability retainer that follows is $1500/month, cancel any month. The watchdog is free either way.
Email me the watchdog.
Drop an email, download the zip right here: watchdog.py, the example config, the Claude Code skills, the tests. The free daily brief rides along; one click kills it.
You're in — grab it below.