caveman-review Skill for Claude: One-Line Code Review Comments

caveman-review makes Claude write code review comments as one line each: where, what is wrong, how to fix it.

Installs: 419.2K; on skills.sh
Stars: 116.7K; on GitHub
Files: 1; besides SKILL.md
Updated: Nov 2026; last repo commit

caveman-review tells Claude to review a diff or pull request with one line per finding in the form L42: problem. fix. Multi-file diffs use file:L42. It can add severity labels for bug, risk, nit and question, and it drops filler such as "I noticed that" and hedging. It triggers on /caveman-review, "review this PR" and "review the diff".

It only writes comments for you to paste. It does not write the fix, approve or request changes, or run linters, and it bundles no scripts. For security findings and architecture disagreements it switches to a full explanation, then goes terse again.

Good for

Developers who want review comments they can scan in seconds
Reviewers on busy pull requests
Authors who want a quick self-review before opening a PR

Install caveman-review

Claude Code

Copy this prompt.

Install the caveman-review skill in Claude Code. Run this command:

npx skills add https://github.com/juliusbrussee/caveman --skill caveman-review

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

From its listing on skills.sh. Source.

What's in the skill

When it loads

Compressed code review - one line per finding with location, problem and fix. Use for /caveman-review, "review this PR", or "review the diff".

Sections in SKILL.md

Rules
Examples
Auto-Clarity
Boundaries

Key facts

Author: JuliusBrussee (repo owner)
License: Apache-2.0
Installs on skills.sh: 419,162 as of 2026-11-02
GitHub stars: 116,702 for the whole juliusbrussee/caveman repo
skills.sh audits: Gen Agent Trust Hub pass, Socket pass, Snyk warn
What is inside: 2 files: SKILL.md and README.md; no scripts
Needs: Nothing beyond Claude and the diff you point it at
Last commit to the repo: 2026-11-01

What it includes

L<line>: problem. fix. One line per finding
Each comment gives the exact line, the problem and a concrete fix. Multi-file diffs use <file>:L<line>.
Try: “Review the diff on this branch with caveman-review.”
Read · Finding format
bug, risk, nit, q labels when findings are mixed
Four optional labels: bug (broken behavior), risk (fragile), nit (style) and q (a real question).
Read · Severity prefixes
No throat-clearing, no hedging, no restating the line
Cuts phrases like "I noticed that" and "you might want to consider". Unsure points become a q: question.
Read · What it drops
Full paragraph for security and design disagreements
Drops terse mode for security findings, architectural disagreements and new authors who need the why, then resumes.
Read · Auto-clarity
Reviews only; no fix, no approval, no linter run
It outputs comments ready to paste into the PR and does nothing else.
Read · Boundaries

From the official docs

Questions

What does the caveman-review skill do?

It makes Claude review a diff with one short line per finding, giving the line number, the problem and the fix. You paste the result into your pull request.

How do I install caveman-review?

Use the install box at the top of this page. In Claude Code, put the skill folder in your personal or project skills folder. In the Claude app, upload it as a skill.

Does it need an API key or other tools?

No. SKILL.md lists no tools, scripts or keys. It is a style guide for review comments, so Claude still needs the diff or files to look at.

Will it fix the code or approve the PR?

No. SKILL.md says it reviews only. It does not write the fix, approve or request changes, or run linters.

Is caveman-review safe?

It bundles no scripts and writes only text. On skills.sh the audits read Gen Agent Trust Hub pass, Socket pass and Snyk warn. As with any review, check each finding before you act on it.

Is a one-line review enough for security problems?

The skill says no. For security findings it switches to a full explanation with a reference, and does the same for architectural disagreements, then returns to short lines.

caveman-review vs a normal Claude code review?

A normal review is prose. caveman-review keeps the same job but limits each comment to location, problem and fix, so there is less to read. Ask for a normal review when you want rationale and alternatives.

Information

updated Nov 2, 2026

Repository: juliusbrussee/caveman
Folder: skills/caveman-review

juliusbrussee/caveman skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 1

caveman-review SKILL.md · github.com