On-call & reliability
Hire an AI on-call l1 responder.
Runs the team's first-response runbook: triages PagerDuty alerts, posts a one-line incident summary, escalates real prod issues, stays quiet on staging warnings.
The job
For every PagerDuty alert, classify severity + environment, post a triage line, escalate real prod incidents.
Held to
Real prod alerts reach a human channel within 60 seconds; staging warnings never page anyone; each escalation includes a recent-deploy note when one exists.
Starts when
An event arrives from your systems.
It works within an allowlist enforced in code — anything outside it pauses and waits for your approval.
Exactly what it can touch
- Incidents reads · Pagerduty
- Incidents acknowledges · Pagerduty
- Post messages · Slack
- Read deployments · GitHub
Proven before it goes live
15 pre-built quality checks — passed before it touches your work.
Every Praxa role ships with an eval suite it must pass before deploying — including the cases where the right answer is to refuse: prompt-injection attempts, out-of-scope requests, malformed input. Deploys that regress against the previous version are blocked. You can re-run the checks any time from the dashboard.
Based on Inspired by PagerDuty Runbook automation · catalog v1.0.0
Or go broader
This role is one sentence of the catalog.
Describe any role in plain English and Praxa builds the employee, proves it the same way, and puts it to work under the same limits.