Overview
Caveman makes an agent answer tersely while keeping code exact, and adds an optional proxy and a compact browse command that shrink what the agent reads.
What it does
- Terse agent replies with exact code
- Optional proxy that trims input tokens
- Compact web pages through caveman browse
- Measured with repeated Claude Code runs
Capability source: Project documentation
Who Caveman fits
Heavy agent users who want lower token use and shorter replies.
How you run it
Caveman runs on infrastructure you manage.
Documented deployment options: npx skills, Proxy.
Before you choose it
Token savings are the project's own measurements. The proxy sits between your agent and the model API, so review what it forwards.
Self-hosting and deployment
- Self-hostable
- Yes
- Cloud / hosted offering
- Not recorded
- Docker noted in deployment tags
- Not recorded
- Deployment tags
- npx skills, Proxy
- Difficulty
- Easy
Caveman questions
- What is Caveman?
- A skill and proxy that make coding agents say less and read less. It is listed under Agent skills on PickOpenSource.
- Is Caveman open source?
- Yes. Its source code is public on GitHub under the Apache-2.0 license.
- Can you self-host Caveman?
- Yes. Listed deployment options: npx skills, Proxy.
- Is Caveman free?
- Pricing is not recorded in this catalog. Check the official website for hosted plans and edition limits, and the Apache-2.0 license for usage conditions. Self-hosting also requires infrastructure and maintenance.
- What are alternatives to Caveman?
- Similar agent skills projects include Superpowers, Skills by Matt Pocock, and Ponytail.
- Is Caveman actively maintained?
- The latest commit on GitHub was October 12, 2026, and the repository has 111k stars.
Related projects
Other approved projects in Agent skills: Superpowers, Skills by Matt Pocock, Ponytail.
Superpowers
A software development method for coding agents, built from composable skills.- Stars
- 297,987
- Updated
- License
- MIT
- Hosting
- Self-hosted
- Category
- Agent Skills
Superpowers
Superpowers gives coding agents a working method: it turns a request into a spec you review, writes an implementation plan, then runs subagent-driven development with review of each task. Its skills trigger automatically once installed.Website GitHubSkills by Matt Pocock
Small, composable engineering skills from Matt Pocock's own agent setup.- Stars
- 286,949
- Updated
- License
- MIT
- Hosting
- Self-hosted
- Category
- Agent Skills
Skills by Matt Pocock
A set of engineering skills Matt Pocock uses daily, written to be small, easy to adapt and composable, and to work with any model instead of owning your whole process.Website GitHubPonytail
A skill that makes coding agents write the smallest change that solves the task.- Stars
- 161,519
- Updated
- License
- MIT
- Hosting
- Self-hosted
- Category
- Agent Skills
Ponytail
Ponytail is one skill prompt that pushes an agent to reuse existing code and native features before adding dependencies, while keeping validation, error handling, security and accessibility. It adds review and repository audit commands.Website GitHubMaintain Caveman? Add the README badge
[](https://pickopensource.site/project/caveman)
Claim this listing to edit its profile, add a screenshot, and see listing analytics.