Guide · Agent skill

How to use the Terse Code Review agent skill (caveman-review)

Updated 2026-10-29 · written with AI from the project's own documentation and checked against our catalogue data

caveman-review is an agent skill that makes an AI agent write code review comments as short, single lines: location, problem, and fix. It is aimed at engineers and teams who want quick, scannable feedback on pull requests and diffs. After this guide you will know what the skill does, how to install it, how to trigger it, and where it falls short.

Terse Code ReviewJuliusBrussee/caveman · Apache-2.0 Download the skill

Quick start

1Download the ZIP. Download the caveman-review ZIP from this page. It contains README.md and SKILL.md.
2Unzip into a skills folder. Unzip into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder.
3Or upload in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the skill.
4Send a review request. Paste the diff or pull request and ask with /caveman-review, "review this PR", or "review the diff".
5Paste the comments. Copy the one-line comments into the pull request. The agent does not post them or approve the PR.

What is an agent skill and what does caveman-review make the agent do?

An agent skill is a folder with a SKILL.md file that tells an AI agent how to handle a specific kind of task. caveman-review tells the agent how to write code review comments when you ask it to review a pull request or code diff.

Each comment is one line that gives the location, the problem, and the fix, with no filler or hedging. The agent can tag each finding by severity: bug, risk, nit, or question. It switches to fuller explanations for security issues, architectural disagreements, and onboarding situations.

Format: L<line>: <problem>. <fix>. or <file>:L<line>: ... for multi-file diffs.
Optional severity prefixes: 🔴 bug, 🟡 risk, 🔵 nit, ❓ q.
It only writes comments ready to paste. It does not change the code, approve the PR, or run linters.

How do you install the caveman-review skill?

Download the caveman-review ZIP from this page and unzip it. The package contains README.md and SKILL.md.

For a personal install, unzip it into ~/.claude/skills/. To limit it to one project, unzip it into that project's .claude/skills/ folder.

In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

Personal: ~/.claude/skills/
Project only: .claude/skills/ inside your repository
Claude apps: Settings → Capabilities → Skills, then upload

How do you trigger caveman-review with a request?

You can call the skill directly with its slash command, or ask in plain language. The skill description lists these triggers: /caveman-review, "review this PR", and "review the diff". The README also lists "code review".

Paste the diff or point the agent at the pull request, then send one of these requests.

/caveman-review
review this PR
review the diff
code review

/caveman-review

What do example outputs look like?

Each finding is one line. Good output keeps exact line numbers and symbol names in backticks, and gives a concrete fix instead of a vague suggestion.

Compare the verbose version with the terse one. The terse comment states the problem and the fix and leaves out the throat-clearing.

Verbose: "I noticed that on line 42 you're not checking if the user object is null..." becomes L42: 🔴 bug: user can be null after .find(). Add guard before .email.
Verbose: "This function is doing a lot of things..." becomes L88-140: 🔵 nit: 50-line fn does 4 things. Extract validate/normalize/persist.
Verbose: "Have you considered what happens if the API returns a 429?" becomes L23: 🟡 risk: no retry on 429. Wrap in withBackoff(3).

L42: 🔴 bug: user can be null after .find(). Add guard before .email.
L88-140: 🔵 nit: 50-line fn does 4 things. Extract validate/normalize/persist.
L23: 🟡 risk: no retry on 429. Wrap in withBackoff(3).
L107: ❓ q

What does the skill need, and what are the safety notes?

The skill lists no requirements. According to the package, it does not run scripts and does not use the network.

Because the skill only reads the diff and writes text, you still need to review its output. It does not run tools or scan code, so it depends on the agent reading the diff correctly.

Requirements: none listed
Runs scripts: no
Uses network: no
Do not rely on it alone for security-critical reviews

What tips help you get better reviews?

Ask for the review on a clear diff or a single pull request so the agent can cite exact line numbers. If you want fuller explanations for a particular comment, say so.

To switch back, say "stop caveman-review" or "normal mode" and the agent reverts to verbose review style.

Give the agent the full diff so line numbers are accurate.
Expect the terse style to omit reasoning; ask for the why when a fix is not obvious.
Use the severity tags to sort comments by priority before posting.

What people use it for

Getting quick, scannable comments on a pull request before a human review.
Producing comments with exact line numbers and symbol names for a terse-feedback team.
Sorting findings by severity using bug, risk, nit, and question tags.
Turning a verbose draft review into one-line comments with concrete fixes.

Frequently asked questions

How do I install the caveman-review skill?

Download the ZIP from this page 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 caveman-review require anything else to run?

The source lists no requirements. It runs no scripts and uses no network access, according to the package details.

Is caveman-review free to use?

The skill is licensed under Apache-2.0. The source does not list prices, so check the licence terms for your own use.

Does caveman-review approve pull requests or fix the code?

No. It only writes review comments ready to paste. It does not change the code, approve or request changes on the PR, or run linters.

Can I use caveman-review for security reviews?

Not on its own. The skill switches to full explanations for security findings, but it does not run tools or scan code, so it depends on the agent reading the diff correctly. Pair it with proper security tooling and human review.

How is caveman-review different from a normal code review prompt?

It forces a consistent format: one line per finding with location, problem, and fix, plus optional severity tags. The terse style can omit reasoning, which the skill adds back only for security, architecture, or onboarding cases.

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

caveman-review suits engineers and teams who want quick, consistently formatted, severity-tagged pull request comments with exact line numbers. It is a poor fit for newcomers who need the reasoning behind every comment, or for security-critical reviews that need real scanning tools.

Download the skill