A new Claude Code skill called 'caveman' by indie developer JuliusBrussee achieved 589 GitHub stars within days of launch and 259 Hacker News points on April 5, 2026. The tool addresses a fundamental cost problem in LLM usage: verbose responses that consume excessive tokens. By stripping filler words, articles, and pleasantries while preserving technical accuracy, caveman reduces token usage by approximately 75% and increases response speed by 3x.
Single-Command Installation Delivers Immediate Cost Savings
The skill installs with a single command: "npx skills add JuliusBrussee/caveman." Users activate it by typing '/caveman' or phrases like 'talk like caveman,' and return to normal mode with 'stop caveman' or 'normal mode.' The system removes unnecessary language elements while maintaining exact technical terms, error messages, and code blocks in their original form.
A concrete example demonstrates the compression: A normal 69-token response stating "The reason your React component is re-rendering is likely because you're creating a new object reference on each render cycle" becomes a 19-token caveman version: "New object ref each render. Inline object prop = new ref = re-render. Wrap in useMemo."
Technical Accuracy Remains Intact Despite Dramatic Compression
The skill modifies Claude Code's system prompts to enforce minimal language while explicitly preserving critical technical elements:
- Code blocks remain in their exact original form
- Error messages are not compressed or modified
- File paths and technical terminology stay 100% accurate
- Debugging information remains fully intact
This ensures that despite the 75% reduction in tokens, developers receive all necessary technical details for implementation. The compression applies only to conversational elements, not to the substantive technical content.
Community Response Highlights Practical Value
Developers on X immediately recognized the practical benefits. One user posted: "I saw that joke about Claude using 75% less tokens when you make it talk like caveman so I made it a one-line install for Claude Code." Another emphasized the value proposition: "Why Use Many Token When Few Token Do Trick? Save 75% tokens and make responses 3x faster."
The GitHub repository's tagline — "🪨 why use many token when few token do trick" — references The Office while perfectly capturing the philosophy: eliminate verbal overhead without sacrificing technical precision.
Cost Reduction Scales With API Usage
The approximately 75% token reduction translates directly to proportional API cost savings. For developers making hundreds or thousands of Claude Code queries, this represents significant cost optimization. The 3x speed increase in response time provides an additional benefit beyond pure cost savings, improving developer workflow efficiency.
JuliusBrussee's background as an indie developer building practical tools for the AI development community positions caveman as a community-driven solution to a widespread pain point. The rapid adoption — nearly 600 stars within days — demonstrates strong product-market fit for token optimization tools.
Key Takeaways
- Caveman skill achieved 589 GitHub stars and 259 Hacker News points within days of launch on April 5, 2026
- The tool reduces Claude Code token usage by approximately 75% while maintaining 100% technical accuracy
- Installation requires a single command: "npx skills add JuliusBrussee/caveman"
- Response speed increases by approximately 3x due to reduced token processing
- Code blocks, error messages, and technical terminology remain in their exact original form despite compression