Skill
Caveman
Token-efficient agent skill — compress output without losing code precision
Install
Ships scriptsnpx skills add JuliusBrussee/caveman
Triggers on
Use when you need to write terser responses while keeping code and commands exact.
- Author
- Julius Brussee
- License
- MIT
Highlights
- Output compression reduces token usage per turn while keeping code, commands, and error messages unaltered
- Switchable intensity modes: lite (minimal compression), full (standard), ultra (aggressive), and wenyan variants for different languages
- Specialized subagent types tune compression for specific workflows — investigator focuses on search, builder on implementation, reviewer on analysis
- Works with 30+ agents through standardized skill installation
- Fractional savings per turn compound over long sessions, particularly useful for large codebases and verbose logs
- Philosophy: eliminate filler, reduce articles, drop hedging language, keep facts exact
How It Works
The skill injects compressed output instructions into agent prompts without changing reasoning. A companion proxy system (runtime component) can further shrink input tokens before API calls. Install via the standard skills package manager and invoke with slash commands or modal switches within your agent interface.
Similar skills
Claude-Mem
SkillCarries what an agent learned in one session into the next.
Agent Engineering
Learning Output Style
SkillInteractive mode that hands the keyboard back at decision points instead of writing everything.
Coding Workflows
Skill Seekers
SkillTurns documentation sites, repos and PDFs into agent skills.
Agent Engineering
Comments(0)
Sign in to comment