Site Reliability agent

The Site Reliability agent writes the operational record. Depending on what it is asked for, it drafts service level objectives (SLOs), alerts, runbook entries or blameless postmortems. In the built-in workflows its job is the postmortem.

Where it runs

It is the fifth and last step of Incident Response. Before it, the Incident Responder triaged the incident, the Architect found the root cause, the Builder described a mitigation and the Tester planned how to verify it.

Inputs

• The session task, with the incident facts you provided.
• A read-only snapshot of relevant files from the project's repository.
• The four earlier steps of the session, in shortened form.

Outputs

A blameless postmortem in text, typically:

• A summary and the impact.
• The timeline from the triage.
• The root cause and the contributing factors.
• What went well and what did not, in detection, response and recovery.
• Follow-up actions, such as an alert that would have caught the issue sooner or a test that would have stopped it.

Because it runs last, the postmortem is shown as the session's summary. The mitigation itself leaves the session as a diff, which goes through the release gate.

Example run

Illustrative example, not a recorded session. After the checkout incident in the example on the Incident Responder page, the Site Reliability agent writes the postmortem. Impact: checkout failed for one payment provider's customers for about 25 minutes. Root cause: a provider response without a currency code, which the new adapter did not handle. Contributing factor: no test used that provider's real response shape. Follow-ups: an alert on checkout error rate above 2% for five minutes, a contract test with recorded provider responses, and a runbook entry on rolling back a deploy.

Limits

• Its timeline is only as accurate as the facts in the task. Check times against your own records.
• It does not create alerts or SLOs in your monitoring system. It drafts them.
• In an incident-response session, the items under the postmortem's Follow-ups heading become Plan board tasks (unless the workspace turned that off). Items it writes elsewhere in the postmortem do not.