Guide · Agent skill

How to use the autoreview agent skill for code review

Updated 2026-11-02 · written with AI from the project's own documentation and checked against our catalogue data

Autoreview is an agent skill that tells an AI coding agent how to run a structured code review on the repository you have open. This page is a pointer: the file in the openclaw/clawhub repository finds or installs the canonical autoreview skill from the openclaw/agent-skills repository, reads its instructions, and then runs its Python helper script. By the end you will know what the skill does, how to install it, how to ask for a review and what to check for safety.

Auto Review (Code Review Pointer)openclaw/clawhub · MIT Download the skill

Quick start

1Check requirements. Confirm you have Python 3, Git and access to the openclaw/agent-skills GitHub repository.
2Install the skill. Unzip the ZIP into ~/.claude/skills/ or a project's .claude/skills/, or upload it under Settings → Capabilities → Skills in the Claude apps.
3Install the canonical skill. If the full skill is missing, run python3 scripts/install-skills autoreview from an openclaw/agent-skills checkout.
4Open the target repository. Start the agent in the repository you want reviewed so the helper runs from its working directory.
5Ask for a review. Request a local or branch review in plain language, stating any engine, model or effort you want.
6Run the helper. Let the agent run the autoreview script with the default Codex command, or the branch variant against origin/main.

What the autoreview skill makes the agent do

An agent skill is a folder with a SKILL.md file that gives an AI agent instructions for a specific task. When the agent is working in a repository, it can read these instructions and follow them.

The autoreview skill asks the agent to review code in the current repository using the canonical autoreview skill and its helper script, scripts/autoreview. This file contains no review logic of its own. It sets defaults for the openclaw/clawhub repository: a Codex engine, the model gpt-6.1-sol, high reasoning effort, a maximum priority of P3, and a local or branch review mode.

It finds or installs the canonical autoreview skill from openclaw/agent-skills.
It reads the installed SKILL.md and the references it requires before reviewing.
It runs the shared scripts/autoreview helper from the repository being reviewed.
It applies this repository's review requirements on top of the canonical skill.

Install the autoreview skill

Download the autoreview ZIP from this page and unzip it into one of these locations: ~/.claude/skills/ for all your projects, or .claude/skills/ inside a single project. In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

The pointer expects the full canonical skill at ~/.agents/skills/autoreview/SKILL.md. If the skill is missing, the agent is told to reuse or clone https://github.com/openclaw/agent-skills and run the install script from that checkout. If you install the canonical skill yourself, use the command below.

Requires Git to clone or update the openclaw/agent-skills checkout.
On Windows, use python instead of python3.
Use --mode copy if symlinks are unavailable on your system.
If the skill is installed at another location you chose, the agent should use that installation instead.

python3 scripts/install-skills autoreview

Trigger the skill with a natural request

Open the repository you want reviewed and ask the agent in plain language to review the code. The agent uses the autoreview skill when the request matches its description, which is structured code review. Keep the request specific about scope so the agent knows whether to review uncommitted local changes or a committed branch.

To pick an engine, model, effort level or service tier yourself, say so in the request. The skill says to preserve any explicit choice you make over its defaults.

Review the local changes in this repository with autoreview.
Run an autoreview of this branch against origin/main.
Use autoreview on this repo with the Codex engine at high reasoning effort.

Run the review command

For the default Codex route, the agent runs the helper from the repository being reviewed and keeps that working directory. The command below is the default for this repository and uses the local review mode.

For a committed branch review, the skill uses --mode branch with --base origin/main instead.

Run it from the root of the repository being reviewed, not from the skill folder.
The helper needs Python 3 and a working Codex engine route, as set by the default command.
Model access outside the file is implied but not specified in the source.

python3 "$HOME/.agents/skills/autoreview/scripts/autoreview" \
  --engine codex --model gpt-6.1-sol --thinking high --codex-speed default \
  --max-priority P3 --mode local

Keep fixes upstream and update the shared checkout

The skill tells the agent to keep shared fixes and instructions in the upstream openclaw/agent-skills repository. Do not add repository-local helpers or test copies to your project.

Update the shared checkout once and every symlinked skill picks up the change. Skills installed in copy mode must be reinstalled after their source is updated.

Fix shared problems in openclaw/agent-skills, not in your project.
Symlinked consumers need only one update of the shared checkout.
Copy-mode skills need reinstalling after the source changes.

Safety notes: scripts and network access

Autoreview runs scripts and uses the network. The agent may clone the openclaw/agent-skills repository and execute its Python scripts. Only use it if you trust that repository and the scripts it contains.

Review the upstream source before you rely on it for code that matters to you.

It runs Python scripts from the installed skill.
It may clone or update a GitHub repository.
Do not use it on code if you have no trust in the external repository.

What people use it for

Running a consistent automated review on local changes before committing to the openclaw/clawhub repository.
Reviewing a committed feature branch against origin/main before opening a pull request.
Reusing one shared openclaw/agent-skills checkout across several projects that link to the same skills.

Frequently asked questions

Is the autoreview skill free to use?

The source file is licensed under MIT. It does not state prices, and the Codex engine route it depends on is not described in terms of cost, so check your own model access arrangements.

What do I need to install autoreview?

You need Python 3, Git, and access to the openclaw/agent-skills GitHub repository. You also need a working Codex engine route, as set by the default command in the file.

Where do I put the autoreview skill folder?

Unzip it into ~/.claude/skills/ for all projects, or into .claude/skills/ inside one project. In the Claude apps you can upload it under Settings → Capabilities → Skills.

Why does the autoreview file say it contains no review implementation?

It is a pointer file. It tells the agent to find the canonical autoreview skill in openclaw/agent-skills, read its instructions, and run its scripts/autoreview helper.

Does autoreview run code on my machine?

Yes. The agent runs Python scripts from the installed skill and may clone or update a GitHub repository. Only use it if you trust the external repository.

Can I use a different model than the default?

Yes. The skill says to preserve an explicit user-selected engine, model, effort or service tier. Name the one you want in your request.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Autoreview suits contributors to the openclaw/clawhub repository and developers who already use the shared OpenClaw agent-skills installation and want consistent automated reviews. It is not a self-contained review tool, since it depends on an external repository, Python, Git and a Codex route. If you do not trust that repository's scripts, or you want a standalone reviewer, pick a different tool.

Download the skill