Your Claude Code scheduled task didn't run. Here's how to know next time.

Why scheduled Claude Code jobs go missing

• Session-bound loops. /loop-style and in-session schedules live as long as the session. A resume, a compaction, or a closed terminal can end them without telling you.
• The host is asleep. Jobs triggered from a laptop's cron or launchd don't run while it sleeps.
• The run "succeeds" but does nothing. The agent hits a permission prompt, a missing tool, or an empty input and exits cleanly.

Add a heartbeat in one line

Because Claude Code can run shell commands, end the task prompt with:

When you're done, run: curl -fsS -m 10 --retry 3 -X POST <your ping URL> -d '{"items": <count>, "text": "<summary>"}'

Create a Did It Run? watch with the schedule you expect (e.g. every day by 7:30am) and a grace window. If the check-in doesn't arrive, you're alerted. Report failures straight away with /fail.

Headless runs from cron or CI

Wrap claude -p in a script that pings /start, sends the output to the ping URL on success, and to /fail on a non-zero exit. The output body powers sanity checks like minimum size or a required keyword.

No shell? Email check-ins are coming soon

If your Claude routine only delivers by email, there's no direct way to check in yet: email check-ins aren't live. For now, relay those emails to the ping URL with an automation tool that watches your inbox (n8n, Make or Zapier), or run the job somewhere that can make an HTTP request.