Operations Playbook
Operations Playbook
The skills on these pages are the ones we actually run against this deployment. They’re not templates, not sketches — they’re the exact Claude Code skill files that get invoked when we ship a new OpenClaw version, review a release, or respond to an incident.
Every phase in every skill exists because of a past failure. The prose on each page calls out why that step is there and which incident taught us we needed it. The embedded skill file is lightly scrubbed for PII (phone numbers, emails, family-specific naming) but otherwise verbatim.
Who this is for
You’re running OpenClaw in production, or about to. You have real users (or family members) depending on the gateway staying up. You’ve already been bitten by at least one upgrade and you’re looking for patterns that aren’t documented in the OpenClaw README.
If you’re evaluating OpenClaw for the first time, start with Getting Started — this section assumes you already have it working.
The skills
| Page | When to use |
|---|---|
| Upgrade Playbook | Before every openclaw update. Covers backup, patch registry, auth health, smoke tests, knowledge refresh. |
More pages are being added — release review, healing, hardening, skill audit, and the Codex performance opt-ins are next.
How to adapt these
Each page ends with a “What to adapt for your setup” section calling out the assumptions that are specific to this deployment (backup script names, patch registry location, custom smoke-test paths). You should be able to lift the skill wholesale, change those assumptions, and run it against your own OpenClaw install.
The skills are Claude Code skills (SKILL.md with YAML frontmatter + markdown body). They live in a Claude Code plugin in this repo; on your side they can live anywhere Claude Code loads skills from.