Deploy Watchdog
helio / deploy-watchdogA 'deploy succeeded' notification isn't the same as a working site. Your AI teammate checks the real thing after every release, then patrols uptime and certs every day after that.
What is Deploy Watchdog?
- ·Confirms a deploy actually changed anything: Fetches key pages and endpoints after every release and checks them against what was supposed to ship — not just the 'deploy succeeded' notice.
- ·Flags stale caching by name: If the old version is still being served, it says so instead of repeating a generic success message.
- ·Runs a daily health patrol: Checks availability, error rate, and certificate expiry once a day, before a customer notices first.
- ·Stays quiet when everything's fine: One DM when something's wrong — no daily noise otherwise.
What does Deploy Watchdog produce?
| ✅Pass/fail deploy confirmation | Every key page/endpoint checked against the expected new version. |
| ⚠️Stale-cache flag | Raised when the old version is still being served after a deploy. |
| 📡Daily availability check | Confirms the service actually responds, not just that the process is running. |
| 🔒Certificate expiry warning | Flagged with enough lead time to renew without a scramble. |
| 🔕Silence when healthy | No notification unless something changed for the worse. |
The SOP, step by step.
- 01
Watch for a deploy event
Triggered by a new build, not on a fixed schedule.
- 02
Fetch key pages and endpoints
The ones that actually reflect whether the new version is live.
- 03
Compare to what changed
Confirm the expected update is actually visible, not just reported as shipped.
- 04
Flag stale caching
If the old version is still being served, say so — don't just repeat 'deploy succeeded'.
- 05
Check availability
Confirm the service actually responds, not just that the process is running.
- 06
Check error rate
Compare against a normal baseline, not an arbitrary fixed number.
- 07
Check certificate expiry
Flag anything expiring soon enough to renew without a scramble.
- 08
Report only on change
A DM when something's wrong — no daily noise when nothing is.
Watch Deploy Watchdog set up a new automation.
What Deploy Watchdog connects to.
Vercel
ConnectedDetects new deploys and checks the live result against what shipped.
GitHub
ConnectedReads build and deploy events to know when a check is due.
Questions
- How is this different from my host's own uptime monitor?
- Most uptime monitors only check that a server responds. This checks that the specific thing you just shipped is actually visible on the page — a subtler and more common failure mode.
- What happens when it finds a problem?
- It reports what it found and, where the fix is well-understood (like a cache purge), can propose or take the fix — always with what changed logged.
- Does it need access to my production environment?
- It only needs the URLs to check and, optionally, GitHub access to correlate failures with specific commits — no production database access required.
- Can it handle a rollback?
- It can prepare a rollback plan and execute one you've approved in advance, but it doesn't decide to roll back unilaterally.
Know your deploy actually worked.
Point Helio at your key pages and endpoints — it verifies every release and patrols uptime every day after.
Ask about this automation