When X happens, run Y. The wiring layer that turns alerts and schedules into actions — restart, page, ticket, webhook, run a recipe. Rate-limited, audited, suspendable, dry-runnable.
php-fpm died, restart php-fpm. If the answer is always the same, waking someone to type it is a process failure, not an incident.
Cooldowns and retry limits are what separate an automation from an outage that restarts itself forever.
When the fix does not hold, the page still happens — with the record that it was tried first.
An alert, a state change or a schedule starts the automation.
A recipe or a service action, behind a cooldown and a retry cap. Without both, a failing restart becomes a restart loop.
When the bounded attempt does not hold, the escalation path is the automation's, not yours to remember.
Automations from Starter; cross-platform actions + complex conditions from Pro.