Summary

This project addresses the high token consumption of AI agents when processing verbose Japanese technical communication. It provides a highly compressed communication skill, genshijin, optimized for Japanese, significantly reducing token usage (up to ~75%) while preserving technical accuracy. This benefits users of AI coding assistants by lowering costs and improving efficiency.

How It Works

Based on the English caveman project, genshijin applies aggressive text compression techniques tailored for Japanese. It removes honorifics, polite language, filler words, cushioning phrases, and redundant expressions, offering three compression levels: Polite, Normal (default), and Extreme. This approach allows AI agents to process more information with fewer tokens.

Quick Start & Requirements

Installation is straightforward via Claude Code's official plugin directory with /plugin install genshijin. Alternatively, it can be installed via npx skills or directly from GitHub. Prerequisites include Python 3.10+ and an Anthropic API key or logged-in claude CLI.

Highlighted Details

• Achieves up to ~75% token reduction with 100% technical accuracy, as demonstrated by extensive benchmarks on Japanese technical tasks.
• Includes specialized sub-skills for generating Conventional Commits (genshijin-commit), providing 1-line PR reviews (genshijin-review), compressing memory files (genshijin-compress), and tracking token usage (genshijin-stats).
• Features automated hooks for session start (rule injection), prompt submission (mode tracking), and statusline badges, ensuring consistent compression.
• Supports integration with multiple AI coding agents beyond Claude Code, including Cursor, Windsurf, and GitHub Copilot, via standardized rule files.
• Offers a standalone installer for environments without plugin support.

Maintenance & Community

No specific details regarding maintainers, sponsorships, or community channels (like Discord/Slack) are provided in the README. The project is an adaptation of the original caveman by Julius Brussee.

Licensing & Compatibility

The project is released under the MIT License, which permits commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

While highly effective for Japanese, benchmarks indicate potential performance degradation or increased token usage in certain English-language scenarios compared to the original caveman or even standard terse output. The system automatically reverts to polite Japanese for security-sensitive operations to prioritize accuracy over compression.