Published skills
genshijin-help
A quick reference card for all genshijin modes, skills, and commands. It's a one-time display with no mode changes or state persistence. Activate it using '/genshijin-help', '原始人ヘルプ', or '原始人の使い方'.
genshijin
Super-compressed communication mode reduces token usage by about 75% by speaking like a caveman, while fully maintaining technical accuracy. It has three intensity levels (Polite, Normal, Extreme) and activates with phrases like 'Caveman Mode' or '/genshijin'.
genshijin-commit
Generates ultra-concise commit messages in Conventional Commits format, prioritizing 'why' over 'what' with subjects up to 50 characters. Supports both Japanese and English, activatable via specific commands or automatically when staging changes.
genshijin-compress
Compresses natural language memory files into a 'Genshijin' format to reduce input tokens, fully preserving technical content, code, URLs, and structure. The compressed version overwrites the original file, saving a human-readable copy as FILE.original.md; activate with '/genshijin-compress <filepath>'.
genshijin-review
Provides super-compressed PR review comments, with one point per line (location, problem, fix), prioritizing signals and supporting Japanese. It can be activated by specific commands or automatically during PR reviews.
genshijin-crew
This guide helps decide when to delegate to Genshijin-style subagents (investigator for code location, builder for 1-2 file edits, reviewer for diff review) instead of inline work or Explore, as their compressed output reduces tool-results by about 60% for longer sessions. Triggers include subagent delegation, genshijin-crew use, agent activation, context saving, and compressed agent output.
genshijin-stats
Displays current session's real token usage and estimated reduction, read directly from Claude Code session logs without AI estimation. Activated by `/genshijin-stats`, output is injected by a mode-tracker hook, not calculated by the model.
Category alert