Guide · Agent skill
How to use the Code Review Assistant agent skill (ce-code-review)
Updated · written with AI from the project's own documentation and checked against our catalogue data
The Code Review Assistant is an agent skill, a packaged set of instructions and helper files that an AI coding agent loads when it needs a specific capability. This guide is for developers who want a structured second review of a change before merging or shipping it. After reading it you will be able to install the skill, ask the agent for a review in plain language, and understand what it reports and what it is allowed to change.
Code Review AssistantEveryInc/compound-engineering-plugin · MIT Download the skill Quick start
- 1Download and unzip. Download the skill ZIP from this page and unzip it to get the skill folder.
- 2Place the skill. Move the folder into ~/.claude/skills/ for personal use or .claude/skills/ inside a project, or upload it under Settings → Capabilities → Skills in the Claude apps.
- 3Confirm git access. Make sure the code you want reviewed sits in a git repository and that git is available.
- 4Ask for a review. Request a review of your current branch or pass a PR link in a plain-language request.
- 5Read the verdict and findings. Check the verdict, the findings and any coverage limits in the report.
- 6Apply only what you choose. Ask the agent to apply findings locally only when you agree with them.
What the skill makes the agent do
When installed, the agent reviews a code change: a named diff, a pull request, or your current branch. It checks the change against its intended behavior and project standards, runs several specialised reviewer passes, and returns a report of real defects and improvements with a clear verdict and any coverage limits.
By default it only reports. It edits your code only when you explicitly ask it to apply the findings locally.
- Looks for bugs, regressions, missing or weak tests, and standards problems
- Reports a verdict and notes any coverage limits
- Does not push, open PRs, or file tickets in any mode
Install the skill from the ZIP
Download the ZIP for this skill from this page and unzip it. Then place the resulting skill folder in one of two locations: your personal skills folder at ~/.claude/skills/, or a project's .claude/skills/ folder so that only that project gets it.
If you use the Claude apps instead, upload the skill under Settings → Capabilities → Skills.
- Personal install: ~/.claude/skills/
- Project install: .claude/skills/ inside the repository
- Claude apps: Settings → Capabilities → Skills, then upload
Check what you need before running a review
The skill needs git, and the directory you review must be a git repository, because the skill uses git rev-parse and diff commands. For pull request links it expects the GitHub CLI (gh) or equivalent access. The source does not specify exact credentials.
Cross-model review is optional. It sends code to an external peer model, and only runs when that route is configured in .compound-engineering/config.yaml.
- git, with the review target inside a repository
- GitHub CLI (gh) or equivalent access for PR links
- Optional: external peer model, set up in .compound-engineering/config.yaml
Ask for a review with a natural request
You do not need special syntax. Ask the agent to review your current branch, a named diff, or a pull request. The skill's argument hint lists the accepted forms: an optional mode:agent token, an optional apply:local token, and either a blank target for the current branch or a PR link.
Passing a PR number, URL or branch name selects what to review. It does not give the agent permission to switch branches. To review uncommitted work, stay on that branch and pass no target or a base: value.
- Review my current branch before I push.
- Review this PR: [paste PR link]
- Review the diff against main and tell me the verdict.
- Review this change and apply the findings locally.
[mode:agent] [apply:local] [blank to review current branch, or provide PR link]Read the report and decide what to apply
The report names what was examined, which review lenses were included, any cross-model pass, and the findings. Each finding carries a suggested class describing the follow-up work it needs. The skill uses these to suggest what to do first, but in report-only runs you decide what to change.
If you want the agent to apply findings, say so explicitly, or include apply:local in the request. A deprecated mode:autofix token does not count as permission.
- Read the verdict first, then the findings
- Check the coverage limits before trusting a clean result
- Apply only the fixes you agree with
Safety notes: scripts and network
This skill runs scripts and uses the network. Cross-model review may send code to an external peer model when that route is configured. Only enable it for code you are comfortable sharing with that model, and do not run unattended reviews until you have checked what would be sent off the machine.
The skill will not check out branches or change your working tree unless you explicitly authorise a local apply.
- Scripts: yes
- Network: yes
- Cross-model: code may leave the machine when configured
- Read-only by default for the working tree
What people use it for
- Reviewing your own branch for bugs and missing tests before pushing
- Getting a structured review receipt on a PR in a team workflow
- Re-checking a change against the project's standards before shipping
- Asking for a report first and then applying selected fixes locally
Frequently asked questions
How do I install the ce-code-review skill?
Download the ZIP from this page and unzip it. Put the folder in ~/.claude/skills/ for personal use or in a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills in the Claude apps.
What do I need to run it?
You need git and a git repository containing the code to review. For PR links the source points to the GitHub CLI (gh) or equivalent access. Cross-model review is optional and needs an external peer model configured in .compound-engineering/config.yaml.
Is the skill free?
The skill is licensed under MIT. The source does not state pricing for the skill itself, so check the repository for any costs tied to models or services you configure.
Does it change my code?
Not by default. It only reports findings. It edits code only when you explicitly ask it to apply findings locally, and it never pushes, opens PRs or files tickets.
Can it review feedback someone already left on my PR?
No. Feedback already left on a PR is handled by the separate ce-resolve-pr-feedback skill, which this skill points to.
Why does the report mention coverage limits?
The skill states what it could not fully check, such as a lens that was not applicable or a peer pass that did not run. Read those notes before treating a clean verdict as complete, especially for very large diffs where a single review may miss issues.
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
Use this skill if you want a structured, report-first second review before merging, or consistent review receipts in a PR workflow. Pick something else if you need to respond to comments already left on a PR, you are reviewing non-code documents or very large diffs, or you cannot allow code to be sent to an external model.
Download the skill