Caveman: a Claude Code skill by Julius Brussee
For developers who want terse commit messages, one-line review comments and lower token spend. Skip it if you prefer full, explanatory prose from Claude.
- Score
- 85of 100
- Rank
- #27of 239
- Installs
- 433Kon skills.sh
- Security
- Clear 4 scanners
- Updated
- 2026
- Context
- ~37tokens always
Install
npx skills add https://github.com/juliusbrussee/caveman Run it in your terminal. View the source
Using the Claude app rather than Claude Code? You can add a skill there too. How to add it in the app
What does Caveman do?
Caveman is a set of Claude skills that make Claude say less. caveman-commit writes a Conventional Commits message with a subject of at most 72 characters and a body only when the why is not obvious, then prints it without running git commit. caveman-review gives one line per finding, such as L42: problem. fix., with optional bug, risk and nit tags. caveman-explore is a read-only search agent that replies only with path:line citations. caveman-learn reads a caveman learn report, shows token sinks, and trims files like CLAUDE.md one edit at a time with your consent. Short rule sets cover refactors, migrations and diagnosing before editing.
It comes as a set of skills that work together:
- caveman-commit (423K installs)
- caveman-review (419K installs)
- caveman-explore (115K installs)
- caveman-learn (114K installs)
- safe-refactor (114K installs)
- investigate-first (113K installs)
- surgical-patch (112K installs)
- lean-build (112K installs)
- caveman-evidence-review (111K installs)
- migration (111K installs)
When should you use Caveman?
Use it when
- You want commit messages in Conventional Commits format
- You want PR comments short enough to paste as is
- Your CLAUDE.md or context costs too many tokens
Skip it when
- You want long, friendly explanations in reviews
- You don't want the caveman CLI the learn skill depends on
What it needs and what it can touch
You need
- The caveman CLI, for caveman-learn and evidence review
What it can do
- Reads project files
- Runs shell commands (caveman learn, caveman mem) in caveman-learn
- Edits files such as CLAUDE.md after you approve each change
From reading its instructions, not from running it.
Why it scores 85
Strongest on upkeep (100 of 100), weakest on trust (65).
Is Caveman safe to install?
Every scanner that checked it passed it. Scanners often miss what other scanners catch, and a skill can change after a scan. Read its source before you install it.
| Scanner | Result | Date |
|---|---|---|
| Gen Agent Trust Hub | Passed | Sep 15, 2026 |
| Socket | Passed | Sep 15, 2026 |
| Snyk | Low risk | Sep 15, 2026 |
| ZeroLeaks | Passed | Apr 15, 2026 |
Details
- Installs
- 433,229
- Repository stars
- 118,478
- Last repository change
- Tokens always loaded
- about 37
- Tokens when used
- about 541
- License
- Apache-2.0
- Data confidence
- High
Did it do the job for you?
Alternatives to Caveman
All skills for this job-
Get git-commitgit-commit
For developers who want Claude to stage and commit changes with Conventional Commits messages.
-
Get code-review-and-qualitycode-review-and-quality
For developers who want a structured review of a diff or pull request before merging.
-
Get security-auditsecurity-audit
For developers who want a source-based security review, from one question up to a full multi-agent audit.
-
Get ponytail-reviewponytail-review
For developers who want a plain-English review of a change that ends in a ship or fix verdict.
Caveman