Guide · Agent skill
How to use the Review Agent agent skill: install and run it
Updated · written with AI from the project's own documentation and checked against our catalogue data
The review-agent skill is a set of instructions that teaches an AI agent to review a code change the way a careful senior engineer would. It is aimed at developers who want a second pair of eyes on a branch or commit before merging, and at teams that delegate review of work-in-progress code to an agent. By the end of this guide you will know what the skill does, how to install it, how to ask for a review in plain language, and what it needs to run safely.
Review Agentpacifio/atlas · Apache-2.0 Download the skill Quick start
- 1Confirm git is installed. The skill runs git diff and git merge-base, so the project must be a git repository and git must be available to the agent.
- 2Download the ZIP. Get the review-agent ZIP from the page where you found this guide.
- 3Unzip the skill. Unzip it into ~/.claude/skills/ for personal use or into the project's .claude/skills/ folder for one repository.
- 4Or upload it in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the skill.
- 5Ask for a review. Describe the target in plain language, such as uncommitted changes, a branch against its base, or a specific commit.
- 6Act on the findings. Work through the findings from P0 down, checking the cited file and line each time.
What the review-agent skill makes the agent do
An agent skill is a folder containing a SKILL.md file of instructions that an agent loads when a task matches the skill's description. This skill is for reviewing code changes. It can review uncommitted edits, a branch diff against its base, a specific commit, or custom review instructions you provide.
The agent reads the diff and enough surrounding code to understand each changed file, checks the relevant tests and call sites, and then reports concrete regressions. Each finding is listed with a priority from P0 to P3 and a file and line reference, with the most severe issues first.
- Reviews uncommitted changes, a base-branch diff, a commit, or custom instructions
- Orders findings by severity rather than producing a long list of style nits
- Cites a file and line range for every finding
- Stays read-only: it is instructed not to edit files, create commits, push branches, post review comments, or hand the review to another agent
Install the review-agent skill
You can install the skill in two ways. Download the ZIP file from the page where you found this guide and unzip it. Then either place the resulting folder in your personal skills directory, ~/.claude/skills/, or in a project's .claude/skills/ directory so that it applies only to that repository.
If you use the Claude apps instead of a local setup, you can upload the skill under Settings → Capabilities → Skills.
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into the project's .claude/skills/ folder
- Claude apps: upload the skill under Settings → Capabilities → Skills
Trigger the skill with natural requests
You do not need to name the skill or use special syntax. The skill's description says it applies when another agent delegates review of uncommitted changes, a base-branch diff, a commit, or custom review instructions, so ask for a review of one of those targets in ordinary language. Be specific about what you want reviewed so the agent picks the right target.
For a base-branch review, the skill compares against the branch's upstream when that upstream exists and is ahead of your local branch. Otherwise it uses the local branch.
- Review my uncommitted changes before I commit.
- Review this branch against its base before I open a merge request.
- Review commit 3f2a1c.
- Review the changes in this branch with focus on the payment code.
What the agent does under the hood
For a base-branch review, the skill is designed to compare the changes that would actually merge rather than diffing directly against the branch tip. It finds the merge base with HEAD and then inspects the diff from that point. If the local branch cannot be resolved, it tries the configured upstream before reporting that the target is unavailable.
The skill only flags an issue when it is introduced by the change, affects correctness, security, performance, or maintainability in a meaningful way, and can be demonstrated from the code. It skips speculative concerns, pre-existing problems and style nits.
git merge-base HEAD <comparison-ref>
git diff <merge-base-sha>Read the findings
Each finding uses a fixed format: a priority, a short imperative title, and a file and line reference. A short paragraph follows explaining the affected scenario and why the behavior is wrong. If nothing qualifies, the skill reports that there are no findings rather than inventing one.
The result ends with a brief overall assessment that may mention test gaps or residual risks.
- P0: universal release blocker or critical failure
- P1: urgent defect that should be fixed next
- P2: ordinary defect that should be fixed
- P3: low-impact issue that is still worth fixing
Safety notes: scripts and network
The skill is documented as running scripts, and it depends on git commands to read your repository. It does not use the network. Because it is read-only by instruction, it will not change your files, but you should still review the skill's contents before installing any skill from a third party.
The skill needs git installed and an agent runtime that can read the repository and run shell commands.
- Runs scripts: yes
- Uses network: no
- Requires git and an agent runtime that can read the repository and run shell commands
What people use it for
- Checking a feature branch against its base before opening a merge request
- Reviewing uncommitted work-in-progress before committing
- Getting a second look at a specific commit a colleague pushed
- Having one agent delegate review of another agent's changes
Frequently asked questions
How do I install the review-agent skill?
Download the ZIP from the page where you found this guide and unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps, you can instead upload it under Settings → Capabilities → Skills.
Does the review-agent skill need git?
Yes. The skill runs git diff and git merge-base, so it only works on projects that use git. It also needs an agent runtime that can read the repository and run shell commands.
Is the review-agent skill free?
The source material provided here does not state pricing or licence terms beyond the Apache-2.0 licence listed for the repository. Check that licence for usage terms.
Does the review-agent skill edit my code or post comments?
No. The skill is instructed not to modify files, create commits, push branches, post review comments, or delegate the review to another agent.
Can the review-agent skill replace human code review?
No. It only flags issues it can demonstrate from the code, so it is not a replacement for human review of security-critical or compliance-sensitive code.
Why doesn't the skill comment on style or formatting?
It is instructed to skip style nits that do not obscure the code, so its findings focus on correctness, security, performance and maintainability.
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 review-agent if you work in a git repository and want an agent to check your branches, commits or uncommitted changes for concrete, severity-ordered defects before you merge. Pick something else if your project does not use git, if you need formal review of security-critical or compliance-sensitive code, or if you want style and formatting feedback.
Download the skill