Runbooks

Runbooks are notebook-style remediation playbooks. Each runbook is an ordered list of typed cells you can read, edit, and run, with a lifecycle that takes a draft all the way to a team standard. In the product, runbooks live under Observability.

Cell types

• markdown: context and documentation for the people running the runbook.
• shell: commands to run.
• http: API calls.

Targets

Each actionable cell can target a specific resource. A cell carries a target type and a small JSON target config that points it at one of:

• a Kubernetes namespace,
• an ECS service, or
• an EC2 instance.

Lifecycle

A runbook moves through four states. It starts as a draft, becomes reviewed once an engineer has checked it, is promoted to standard when it becomes a team-blessed playbook, and is marked deprecated when it is obsolete.

Running a runbook

You can run a runbook and see its run history, with per-cell output captured for each run. This gives you a record of what was executed and what each step returned.

Authoring with AI

A "Create with AI" flow has Claude draft the runbook and generate shell or Python scripts for its cells. An engineer then reviews the draft before it is trusted. You can also author runbooks from your AI assistant through the MCP server.

Permissions

• Viewing a runbook requires observability.view.
• Editing a runbook requires observability.manage.
• Running a runbook requires observability.runbook.execute.

What is next

• MCP server covers authoring runbooks from Claude through the Model Context Protocol.