@preapexis/pi-kit
Personal Pi coding-agent kit with safety extensions, status UI, prompt workflows, skills, and themes.
Package details
Install @preapexis/pi-kit from npm and Pi will load the resources declared by the package manifest.
$ pi install npm:@preapexis/pi-kit- Package
@preapexis/pi-kit- Version
2.3.5- Published
- Jul 19, 2026
- Downloads
- 6,087/mo · 428/wk
- Author
- preapexis
- License
- ISC
- Types
- extension, skill, theme, prompt
- Size
- 152.7 KB
- Dependencies
- 0 dependencies · 3 peers
Pi manifest JSON
{
"extensions": [
"./extensions"
],
"skills": [
"./skills"
],
"prompts": [
"./prompts"
],
"themes": [
"./themes"
]
}Security note
Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.
README
Pi Package with Caveman Skills
This package demonstrates how to add third-party skills to the pi system. It includes the Caveman skill set from https://github.com/juliusbrussee/caveman.
Overview
Caveman is an ultra-compressed communication mode that cuts output tokens by 65% while preserving full technical accuracy. It makes AI coding agents talk like cavemen, keeping code, commands, and errors byte-for-byte exact.
Added Skills
This package adds the following skills from caveman:
- caveman - The main skill for compressed communication mode
- caveman-commit - Ultra-compressed commit message generator
- caveman-compress - Rewrite memory files into caveman-speak
- caveman-review - One-line PR comments for bug detection
- caveman-stats - Session token usage and savings tracking
- cavecrew - Subagents for investigation and review
How to Use
Once installed in your agent:
- Say "talk like caveman" or use
/cavemanto enable compressed mode - Use
/caveman-statsto see token savings - Use
/caveman-commitfor conventional commit messages - Use
/caveman-reviewfor PR analysis - Use
/caveman-compress <file>to compress memory files
Files Added
skills/- The six caveman skillsextensions/- Permission modes extensionREADME.md- This documentation
Testing
To test these skills, you would need to:
- Install them in your AI agent
- Enable caveman mode
- Test various commands and see how they compress output