On-Call Agent: PagerDuty Incident Triage with Approval-Gated Remediation

Turns a raw PagerDuty page into a triaged incident brief. The agent gathers signal from Datadog and GitHub, reasons about the likely cause, and proposes concrete fixes — but never acts on its own. A human approves in Slack first.

Use it when your on-call rotation gets paged faster than humans can context-switch, and you want the first five minutes of investigation done before anyone opens a laptop. Best for teams that want speed without handing an agent unattended write access to production.

1. 1A PagerDuty incident fires the workflow with the affected service and severity.
2. 2The agent queries Datadog for that service's error rate, latency, and saturation over the last hour.
3. 3It pulls the service's recent merged commits and deploys from GitHub to spot suspicious changes.
4. 4Logic ranks candidate causes and matches each to a runbook remediation (rollback, scale up, restart, feature-flag off).
5. 5The agent posts a diagnosis plus an ordered list of proposed steps to the incident's Slack channel, each behind an Approve button — nothing executes until a responder clicks.