Alarm checks
Some checks are built to fail when something in the real world needs attention. An example is a GitHub Actions workflow that fails on purpose when image coverage drops under a floor.
Mark that check as an alarm. Then a failure reads as an alarm, not "deploy failed". Any fix job is told to fix what the check measures, and never to weaken the check.
You can mark a GitHub Actions workflow, or a check-in job, on a project.
Mark one
Marking needs a write key. The check is the workflow name, the workflow file, or the job name. note is optional and says what it watches.
bash
printf 'Authorization: Bearer %s\n' "$SPOOKY_API_KEY" | curl -s -X POST -H @- -H "content-type: application/json" \
-d '{"project":"owner/repo","check":"Watch visual coverage","note":"image coverage floor"}' \
https://spookyparticles.com/api/v1/alarmsList them
Listing needs a read key.
bash
printf 'Authorization: Bearer %s\n' "$SPOOKY_API_KEY" | curl -s -H @- https://spookyparticles.com/api/v1/alarmsUnmark one
Send the same project and check with DELETE.
bash
printf 'Authorization: Bearer %s\n' "$SPOOKY_API_KEY" | curl -s -X DELETE -H @- -H "content-type: application/json" \
-d '{"project":"owner/repo","check":"Watch visual coverage"}' \
https://spookyparticles.com/api/v1/alarmsEvery change is recorded in the audit log.
MCP tools
Your agent can do the same over MCP. See MCP server.
| Tool | Key | What it does |
|---|---|---|
list_alarms | read | Lists alarm checks. |
mark_alarm | write | Marks a check as an alarm. |
unmark_alarm | write | Clears the mark. |