genshijin (InterfaceX): a Claude skill that makes replies in Japanese short, like a caveman, to save tokens

genshijin makes Claude answer in clipped Japanese that drops polite filler, so replies use fewer tokens while code and technical terms stay intact.

Installs: 12.3K; on skills.sh
Stars: 331; on GitHub
Updated: Oct 2026; last repo commit

Type /genshijin or say 「原始人モード」, 「短く」 or 「トークン節約」 and Claude cuts honorifics, cushion words, hedges and openers from its Japanese replies. There are three levels: 丁寧 (polite but short), 通常 (the default) and 極限 (keywords only). The skill says it keeps technical content exact and cuts about 75% of the tokens. Say 「原始人やめて」 or 「通常モード」 to turn it off.

It is a Japanese version of the English caveman skill. SKILL.md keeps code blocks, error messages, numbers and negations unchanged. It switches back to normal Japanese for destructive actions and security warnings. The skill file alone bundles no scripts. The plugin route adds hooks, extra skills and commands, described below.

Good for

Japanese-speaking developers who want shorter answers from Claude
Heavy Claude Code users who watch token use and cost
Teams who want the style to hold across a long session

Install genshijin

Claude Code

Copy this prompt.

Install the genshijin skill in Claude Code. Run this command:

/plugin install genshijin

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

The plugin adds two hooks, six more skills, three subagents and slash commands such as /genshijin-commit and /genshijin-stats. Pick one route. The README also describes copying only SKILL.md into ~/.claude/skills/genshijin for a skill with no hooks.

From the project README. Source.

What's in the skill

When it loads

超圧縮コミュニケーションモード。原始人のように話してトークン使用量を約75%削減。技術的正確性は完全に維持。強度レベル: 丁寧・通常（デフォルト）・極限の3段階。「原始人モード」「短く」「簡潔に」「トークン節約」と言うか、/genshijin で起動。

Sections in SKILL.md

永続性
ルール
強度レベル
自動解除 (Auto-Clarity)
境界
テキスト形式ファイル生成

Key facts

Author: interfacex-co-jp (GitHub)
Repo: interfacex-co-jp/genshijin (skills/genshijin)
License: MIT
Installs on skills.sh: 12,280 as of 2026-11-02
GitHub stars: 331 (whole repo) as of 2026-11-02
Last repo commit: 2026-10-11
Folder contents: 1 file in the skill folder: SKILL.md. The full plugin repo adds 6 more skills, 3 subagents, 6 slash commands and hooks
Plugin version: 1.5.0 (plugin.json, MIT)
skills.sh audits: Gen Agent Trust Hub pass, Socket pass, Snyk pass

What it includes

Turns the mode on; add 丁寧, 通常 or 極限 to choose the level.
/genshijin
Removes honorifics, cushion words, hedges, openers, padding and repeated meaning.
Deletion rules
Code, error text, numbers, technical terms, negations and the user's main language.
What it keeps
Returns to normal Japanese for destructive actions, security warnings and confusion.
Auto-clarity
Asks once before writing .md, config and data files in caveman style.
File generation check
Holds the level every turn until you say 「原始人やめて」 or 「通常モード」.
Persistence

From the official docs

Permissions

Medium risk

The skill file only changes how Claude writes, but the plugin route installs hooks that run Node scripts on every session and prompt. The optional compress command rewrites memory files and sends their text to Claude.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

The plugin runs hooks on every session and prompt
plugin.json runs genshijin-activate.js at SessionStart and genshijin-mode-tracker.js at UserPromptSubmit, both with node. The activate script reads ~/.claude/settings.json to check the status line. Skip the plugin and use the skills CLI if you want only SKILL.md.
/genshijin-compress rewrites your memory files
It compresses a file such as CLAUDE.md in place, saves the original as a .original.md backup and needs Python 3.10+ plus ANTHROPIC_API_KEY or a logged-in claude CLI. The README says it refuses .env, credentials and .ssh paths.
/genshijin-stats reads your session transcripts
The prompt hook runs genshijin-stats.js, which reads transcripts under ~/.claude/projects and keeps a history file at ~/.claude/.genshijin-history.jsonl.
Optional installers change your setup
The README also offers curl-to-bash installers for several agents and a standalone hooks installer that merges into settings.json. Read them first.

Technical details

Talks to	Anthropic API (only for /genshijin-compress, via ANTHROPIC_API_KEY or the claude CLI)

Questions

What does genshijin do?

It makes Claude reply in compressed Japanese. It removes honorifics, filler, hedging and padding, and keeps code, error messages, numbers and negations as they are. SKILL.md says it keeps technical accuracy and cuts about 75% of tokens. That is the author's claim.

How do I install it in Claude Code?

The install box shows the plugin route first. It installs the full plugin from the repo, which adds hooks that keep the style from drifting, plus six more skills, three subagents and slash commands. The skills CLI and manual routes give you the skill without those extras.

What do the plugin hooks do?

The plugin runs a Node script at session start that injects the active rule into the chat. A second script runs on each prompt to track mode changes and add a short reminder. The /genshijin-stats command reads your session transcript to show token use. The README also describes an optional status line badge.

What are the three levels?

丁寧 keeps polite speech but drops filler and hedges. 通常 is the default and drops honorifics and many particles. 極限 drops grammar and uses keywords, arrows and abbreviations. Switch with /genshijin 丁寧, /genshijin 通常 or /genshijin 極限.

Does it work in the Claude app or Codex?

The README says it targets Claude Code and Codex. The Claude app ZIP upload in the install box carries only the skill file, so the plugin hooks do not apply there. Codex installs through the skills CLI.

Does it need an API key?

The skill itself does not. The separate /genshijin-compress command in the plugin needs Python 3.10+ and either ANTHROPIC_API_KEY or a logged-in claude CLI, according to the README.

Is it safe?

The skill file only changes how Claude writes. The plugin runs Node hooks on every session and prompt, and /genshijin-compress rewrites memory files such as CLAUDE.md. The README says it keeps a backup and refuses .env, credentials and .ssh files. Read the hooks before you install the plugin.

What is the difference from caveman?

genshijin is a Japanese version based on the English caveman skill. The README says it targets Japanese-specific padding such as honorifics and cushion words instead of English articles and filler.

Information

updated Nov 2, 2026

Repository: interfacex-co-jp/genshijin
Folder: skills/genshijin

interfacex-co-jp/genshijin skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 4

genshijin README · github.com
genshijin plugin.json · github.com
genshijin SKILL.md · github.com
genshijin on skills.sh · skills.sh