Back to catalog / Agent workflows
Investigate an SRE alert and request rollback approval
Start from an alert, correlate telemetry, GitHub changes, infrastructure, and prior incidents, then report evidence in Slack and await responder approval without deploying.
Goal
Start from an alert, correlate telemetry, GitHub changes, infrastructure, and prior incidents, then report evidence in Slack and await responder approval without deploying.
Steps
- Install the bot in test Slack and send a synthetic alert with a unique fingerprint. [src-openai-triage-sre-alert-and-request-rollback-approval]
- Read sample metrics, deployments/commits, and past incidents to create a sourced investigation thread. [src-openai-triage-sre-alert-and-request-rollback-approval]
- Record approve/decline in test Slack and confirm no deployment ran. [src-openai-triage-sre-alert-and-request-rollback-approval]
How to verify
Verify the thread includes the alert fingerprint and supporting evidence; approval may be recorded but deployment state remains unchanged.
Troubleshooting
If duplicate threads appear, inspect fingerprint deduplication; if evidence is missing, check fixtures and read-only integration scopes.
Permissions
Prerequisites
- Python 3.14+, Docker sandbox, OpenAI API and executor keys, test Slack app, and synthetic alerts.
Permissions
- Posts Slack messages and reads telemetry/GitHub/AWS; it records rollback approval but never deploys, and the executor key must be restricted.
Limitations
Bundled sample data is not production incident evidence; authorize and verify alert sources before connecting live systems.
Pinned upstream source
- Repository
- openai/openai-cookbook
- File path
- examples/agents_api/apps/sev_bot/README.md
- Pinned commit
5bd6cb40b6b84b69b2032d810610aa85064102df- Source checked
- 2026-10-14 · Reviewed against the pinned upstream source
Repository metrics
These values describe the upstream repository, not popularity or likes for an individual skill or playbook.
- Stars
- Not collected
- Forks
- Not collected
- Observed at
- Not collected
No metrics snapshot is available yet.
License: MIT
Main topic
Make Agent behavior safer and easier to verify