Give Codex and Claude Code the context to set up, validate, and troubleshoot your OpenClaw installation.
Bring your coding agent to your local OpenClaw install, describe what you want, and let it do the heavy lifting:
- create or update skills
- wire providers and integrations
- add cron/scheduled automation
- inspect config safely
- validate changes
- troubleshoot broken installs
OpenClaw is powerful, but setup can still be intimidating.
A lot of the friction is not “writing code” — it’s knowing:
- where the right files live
- which commands to run first
- how to validate changes safely
- how to structure skills and supporting files
- how to troubleshoot without random trial-and-error
OpenClaw Operator gives coding agents like Codex and Claude Code the missing operational context.
Instead of manually piecing together docs, configs, skills, cron jobs, and validation steps, you can point your coding agent at your OpenClaw setup and say what you want.
This repo provides:
- a reusable
SKILL.mdfor OpenClaw operations - a
validation_checklist.mdfor safe completion and verification - a
task-playbooks.mdreference for common workflows - a lightweight
AGENTS.mdfor Codex - a lightweight
CLAUDE.mdfor Claude Code - a structure you can copy into your own repo or local OpenClaw install
This is for you if you:
- run OpenClaw locally
- want to use Codex or Claude Code to configure it
- want a repeatable structure for OpenClaw setup work
- are tired of re-explaining OpenClaw conventions to your coding agent every session
This is not a replacement for OpenClaw itself.
It does not magically install or configure every environment without review.
It is an operator pack: a set of instructions, playbooks, and validation habits that help a coding agent work on an OpenClaw installation more effectively.
Inspect this OpenClaw installation, identify why the provider config is failing, fix it with the smallest safe change, and validate the result before finishing.
With this pack installed, your coding agent should be better at tasks like:
- “Create a new OpenClaw skill for my local media workflow.”
- “Add a scheduled job and validate that it’s wired correctly.”
- “Inspect this install and tell me why the gateway/channel is failing.”
- “Set up provider config safely and confirm the health checks pass.”
You can use this pack in three places:
- Codex
- Claude Code
- OpenClaw itself
You can install one, two, or all three depending on your workflow.
Codex reads project guidance from AGENTS.md and scans skills from .agents/skills/.
From the root of your repo:
mkdir -p .agents/skills
cp AGENTS.md ./AGENTS.md
cp -R skills/openclaw-operator .agents/skills/openclaw-operatorThen open the repo in Codex and prompt it with an OpenClaw setup task.
Read AGENTS.md and the openclaw-operator skill, inspect this OpenClaw setup, and help me add a new cron job with validation.
Claude Code can use CLAUDE.md for persistent repo guidance and .claude/skills/ for skills.
From the root of your repo:
mkdir -p .claude/skills
cp CLAUDE.md ./.claude/CLAUDE.md
cp -R skills/openclaw-operator .claude/skills/openclaw-operatorUse the OpenClaw Operator skill to inspect this installation, fix the provider configuration with the smallest safe change, and validate it before finishing.
OpenClaw loads skills from either:
<workspace>/skills~/.openclaw/skills
You can install this skill into either location.
mkdir -p ./skills
cp -R skills/openclaw-operator ./skills/openclaw-operatormkdir -p ~/.openclaw/skills
cp -R skills/openclaw-operator ~/.openclaw/skills/openclaw-operatorThe best setup for most people is:
- put
AGENTS.mdin the repo root - put
CLAUDE.mdin.claude/CLAUDE.md - install the skill into:
.agents/skills/openclaw-operator.claude/skills/openclaw-operator- optionally
~/.openclaw/skills/openclaw-operator
That way:
- Codex can use it
- Claude Code can use it
- OpenClaw-native agents can use it too
The core idea is simple:
- your coding agent sees an OpenClaw-related task
- it consults the
openclaw-operatorskill - it inspects the current state before editing
- it makes the smallest safe change
- it validates the result before declaring success
This pack is opinionated in a few useful ways:
- inspect before editing
- prefer official CLI/docs over memory
- keep changes narrow and reversible
- preserve existing structure and naming
- validate before finishing
- never leak secrets in logs, diffs, or files
The main operating playbook.
Use it for:
- when to activate the skill
- how to inspect an OpenClaw environment
- how to approach edits safely
- how to report back
The file that helps a coding agent verify work before saying “done.”
Use it for:
- health checks
- skill validation
- config validation
- channel/integration checks
- security/hygiene checks
- completion standards
The file for common OpenClaw workflows.
Use it for:
- creating skills
- setting up cron/scheduled automation
- wiring providers
- troubleshooting installs
- common first-pass inspection flows
A place for reusable examples, templates, snippets, and scaffolds.
This is intentionally lightweight in v1, but it is designed to grow over time.
Read the OpenClaw Operator skill and create a new custom OpenClaw skill for managing my home server backups. Keep the skill concise, add any supporting references if needed, and validate the result.
Use the OpenClaw Operator skill to add a scheduled OpenClaw task that runs every morning, summarize what files you changed, and validate that the schedule is wired correctly.
Inspect this OpenClaw installation, determine why it is failing, identify the most likely root cause, fix it with the smallest safe change, and show exactly how you validated the fix.
Use the OpenClaw Operator skill to inspect the current provider configuration, fix any obvious issues, and validate gateway health without exposing secrets.
This repo is designed to help coding agents work more safely, not more recklessly.
Please review sensitive changes carefully, especially when tasks involve:
- API keys
- auth flows
- channels/integrations
- system services
- scheduled jobs
- workspace-wide configuration
A coding agent should still be treated like an operator that needs review, not blind trust.
Here’s the workflow I recommend:
- clone this repo or copy the pack into your own repo
- install the skill for the coding agent you use
- open your real OpenClaw install/workspace in Codex or Claude Code
- describe what you want in plain English
- let the agent inspect the current state first
- review the proposed changes
- make sure it validates before finishing
No.
Minimum useful setup:
skills/openclaw-operator/SKILL.md
skills/openclaw-operator/references/validation_checklist.md
Recommended setup:
AGENTS.md
CLAUDE.md
skills/openclaw-operator/SKILL.md
skills/openclaw-operator/references/validation_checklist.md
skills/openclaw-operator/references/task-playbooks.md
Not necessarily.
If your goal is to use Codex or Claude Code to work on an OpenClaw install, the most important thing is to install the pack where your coding agent can see it.
Installing it inside OpenClaw is optional, but useful if you want OpenClaw-native agents to have the same skill available.
No.
It is especially helpful for people who know what they want to accomplish but do not want to memorize OpenClaw conventions, file locations, and validation steps.
No.
This is a community-created operator pack built to help coding agents work more effectively with OpenClaw.
Planned improvements:
- starter assets and scaffolds
- more example prompts
- better screenshots/GIFs
- a short demo video
- a ClawHub-friendly distribution path
- more playbooks for integrations and troubleshooting
PRs and issues are welcome.
Good contributions include:
- better task playbooks
- safer validation flows
- clearer install instructions
- more realistic demo prompts
- improvements to the README visuals
- tested examples for real OpenClaw workflows
MIT
Built for the OpenClaw community and anyone who wants to turn OpenClaw setup into a more self-serve workflow with coding agents.
If this repo helps you, star it, share it, and send improvements back upstream.