genshijin-crew Skill for Claude Code: Subagents With Short Replies

Tells Claude Code when to hand work to three short-reply subagents, so long sessions keep more room in the main context. Needs the genshijin plugin.

Installs: 11.9K; on skills.sh
Stars: 331; on GitHub
Updated: Oct 2026; last repo commit

genshijin-crew is a decision guide. It says when to send a job to one of three subagents (genshijin-investigator, genshijin-builder, genshijin-reviewer) instead of working in the main thread or using the built-in Explore agent. The subagents return compressed output, and the SKILL.md claims this cuts the tool results that come back to the main context by about 60%.

The skill is Japanese text, and it does not contain the subagents. They live in the genshijin plugin's agents/ folder, so install the plugin and not this folder alone. The README says the skill triggers automatically.

Good for

Claude Code users who run long sessions and hit context limits
Developers who want fast code lookups and small diffs reviewed in one-line findings
People already using the genshijin short-reply mode

Install genshijin-crew

Claude Code

Copy this prompt.

Install the genshijin-crew 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.

Install the whole plugin: the 3 subagents this skill points to are plugin files. After install, Claude Code can start genshijin-investigator, genshijin-builder and genshijin-reviewer.

From the project README. Source.

What's in the skill

When it loads

原始人スタイル subagent への委譲判断ガイド。`genshijin-investigator` (コード位置特定)、 `genshijin-builder` (1-2ファイル編集)、`genshijin-reviewer` (diff レビュー) を inline作業 or vanilla `Explore` の代わりにスポーンするタイミングを示す。subagent 出力は原始人圧縮 → 主コンテキストに戻る tool-result が約60%縮小 → 長セッション持続。 Trigger: 「subagent 委譲」「genshijin-crew 使用」「investigator/builder/reviewer 起動」「コンテキスト節約」「圧縮 agent 出力」。

Sections in SKILL.md

genshijin-crew vs 代替の使い分け
なぜ存在するか (実利)
出力契約
チェイニングパターン
禁止事項
自動解除 (継承)

Key facts

Author: InterfaceX-co-jp
Repo: interfacex-co-jp/genshijin
License: MIT
Installs on skills.sh: 11,912 (Nov 2, 2026)
GitHub stars: 331 (Nov 2, 2026), whole repo
Last commit: Oct 11, 2026 (repo)
Files in the skill folder: 1 file including SKILL.md
Needs: The 3 subagents in the genshijin plugin's agents/ folder
Starts: Automatically, per the README

What it includes

Locates code and returns a file:line list.
Answers "where is X defined", "who calls Y" and "every use of Z". The agent file calls it read-only and says it refuses to propose fixes, but its tools are Read, Grep, Glob and Bash, on the haiku model.
Runs code · genshijin-investigator
Makes small edits in 1 or 2 files and replies with a diff receipt.
The agent file gives it Read, Edit, Write, Grep and Glob, no Bash. It refuses jobs touching 3 or more files with `too-big.`.
Write · genshijin-builder
Reviews a diff, branch or file with one finding per line.
Each line is `path:line: severity: problem. fix.` and ends with totals. The agent file gives it Read, Grep and Bash and the haiku model.
Runs code · genshijin-reviewer
Maps task types to the right agent, or to the main thread or vanilla Explore.
The SKILL.md table sends architecture explanations to `Explore`, features touching 3+ files to the main thread, and deep reviews to `Code Reviewer`.
When-to-delegate table
Locate, then fix, then verify; parallel scouts; single edits.
Run the investigator, give 1 or 2 sites to the builder, then have the reviewer audit the diff. For wide searches, start 2 or 3 investigators in one message.
Chaining patterns
Lists 4 things to avoid, such as using the builder before the files are known.
Other rules: no investigator-to-builder chain for a 5-file refactor, no general-feedback requests to the reviewer, and expect structured output that may be cryptic.
Do-not rules

From the official docs

Permissions

Medium risk

The subagents behind this skill can run shell commands (investigator, reviewer) and edit files (builder), and the plugin adds hooks on every prompt. Check the agent files and plugin.json before you install.

What it can do

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

Watch out for

Two agents can run shell commands
genshijin-investigator and genshijin-reviewer list Bash in their tools. Work with your usual permission prompts on. github.com
The builder edits and writes files
genshijin-builder has Edit and Write but no Bash, and limits itself to 1-2 files. Review its diff receipt. github.com
The plugin adds prompt hooks
plugin.json registers Node scripts on SessionStart and UserPromptSubmit. The SessionStart script reads ~/.claude/settings.json to check the status line. Skim them first. github.com

Technical details

Scopes	Subagent tools: Read, Grep, Glob, Bash, Edit, Write Plugin hooks on SessionStart and UserPromptSubmit

Questions

What does genshijin-crew do?

It guides Claude Code on when to delegate to three short-reply subagents: an investigator for finding code, a builder for 1-2 file edits, and a reviewer for diffs. It does not do the work itself.

How do I install it?

Use the genshijin plugin from the install box above. The skill is only the delegation guide, and the three agent files ship in the plugin.

Does it work without the plugin?

The skill folder alone gives Claude the guide but not the agents it names. The README does not say the skills CLI installs the agents, so use the plugin.

Are the replies in Japanese?

The SKILL.md and the agent prompts are written in Japanese, and the README describes the style as compressed Japanese. Read the agent files if you need to know exactly how they reply.

Is it safe?

The investigator and reviewer agents can run Bash, and the builder can edit and write files in your project. The plugin also adds prompt hooks. See the permissions section, and read the agent files before installing.

genshijin-crew vs Explore?

Per the SKILL.md, use the investigator when you want only locations and a compact answer. Use vanilla Explore when you also want architecture explanations or suggestions.

Information

updated Nov 2, 2026

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

interfacex-co-jp/genshijin skills.sh

Sources

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

Show all 7

genshijin README · github.com
genshijin plugin.json · github.com
genshijin-crew SKILL.md · github.com
genshijin-investigator agent file · github.com
genshijin-builder agent file · github.com
genshijin-reviewer agent file · github.com
genshijin-activate.js · github.com