PluginBench
Skill
Pass
Audit score 90

genshijin-crew

interfacex-co-jp/genshijin

Delegate code investigation, editing, and review to compressed subagents to reduce context consumption by ~60%.

What is genshijin-crew?

genshijin-crew provides three specialized subagents—investigator, builder, and reviewer—that output in a compact "primitive" format instead of prose. Use this when you need to preserve context tokens across many delegations in long coding sessions, trading verbosity for efficiency.

  • Spawn genshijin-investigator to locate definitions, callers, and usage patterns across files with structured path:line output
  • Spawn genshijin-builder for surgical edits to 1–2 files with verification and change summaries
  • Spawn genshijin-reviewer to audit diffs and branches with severity-tagged findings
  • Chain investigators → builders → reviewers in sequence or parallel for multi-step workflows
  • Compress subagent tool-results to ~1/3 token size, reducing main-thread context consumption per delegation

How to install genshijin-crew

npx skills add https://github.com/interfacex-co-jp/genshijin --skill genshijin-crew
Claude Code
Cursor
Windsurf
Cline

How to use genshijin-crew

  1. 1.Identify the task: location lookup → use genshijin-investigator; surgical edit (1–2 files, clear scope) → use genshijin-builder; diff review → use genshijin-reviewer
  2. 2.For investigator: provide a symbol or pattern to search; output will be path:line — symbol — note format
  3. 3.For builder: provide specific file paths and line ranges; output will confirm changes or return terminal status (too-big, needs-confirm, ambiguous, regressed)
  4. 4.For reviewer: provide file or branch to audit; output will be path:line: emoji severity: issue. fix. format with totals
  5. 5.Chain subagents: investigator → builder → reviewer for full workflows, or run multiple investigators in parallel for broad scans

Use cases

Good for
  • Find all references to a symbol across a codebase, then edit the 2 most critical call sites and review the diff
  • Investigate where a bug originates (investigator), fix it in one file (builder), and validate the change (reviewer)
  • Scout a large refactor in parallel: run investigator on definitions, callers, and tests simultaneously, then aggregate findings in main thread
  • Preserve context in long sessions by using compressed subagents instead of vanilla Explore for routine code lookups
Who it's for
  • Developers working in long coding sessions who need to manage context budget
  • Teams performing multi-step code investigations and edits
  • Anyone using Claude Code or Cursor who wants to delegate without exhausting token limits

genshijin-crew FAQ

When should I use genshijin-crew instead of vanilla Explore or Code Reviewer?

Use genshijin-crew when you need compressed output to save context tokens—typically for routine code location, surgical edits, or diff audits. Use vanilla agents when you need prose explanations, architectural rationale, or deep alternatives analysis.

Can I use genshijin-builder without running investigator first?

Only if you already know the exact file paths and line ranges. Using builder without pre-identified locations wastes tokens because the main thread must provide context. Always run investigator first for unknown code.

What does 'too-big' or 'regressed' mean in builder output?

'too-big' means the edit scope exceeds 2 files or is too complex for builder; escalate to main thread or feature-dev:code-architect. 'regressed' means the change introduced a new issue; builder will not apply it.

Can I ask genshijin-reviewer for architectural feedback or alternatives?

No. Reviewer returns findings only (bugs, issues, severity tags). For architectural opinions and alternatives, use vanilla Code Reviewer or main thread discussion.

How much context do I actually save?

Typical vanilla Explore returns ~2000 tokens of prose; genshijin-investigator returns ~700 tokens of structured output for the same discovery. Over 20 delegations in a session, this difference determines context exhaustion vs. task completion.

Full instructions (SKILL.md)

Source of truth, from interfacex-co-jp/genshijin.


name: genshijin-crew description: > 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」。

genshijin-crew = 原始人形式で出力する3 subagent preset。役割は Anthropic デフォルト (Explore、編集系 agent、reviewer) と同じ。差分は返ってくる tool-result が圧縮済 → 主コンテキスト消費が委譲毎に縮む。

genshijin-crew vs 代替の使い分け

タスク使用
「Xの定義どこ / Yを呼ぶ箇所 / Zの全用法」genshijin-investigator
同上 + アーキテクチャ解説/提案も欲しいExplore (vanilla)
Surgical編集、≤2ファイル、スコープ明確genshijin-builder
新機能 / 3+ファイル / cross-cutting refactor主スレッド or feature-dev:code-architect
Diff/branch/file の bug レビューgenshijin-reviewer
rationale + alternatives 付き深いコードレビューCode Reviewer (vanilla)
1行回答済の確実な内容主スレッド、subagent不要

判断基準: subagent 出力を1/3トークンで欲しいなら genshijin-crew、散文で欲しいなら vanilla。

なぜ存在するか (実利)

Subagent tool-result は主コンテキストに verbatim 注入される。Vanilla Explore が散文2k tokens 返すと毎回主コンテキスト2k消費。同じ発見が genshijin-investigator だと約700 tokens。1セッション20委譲で context exhaustion vs タスク完了の差。

出力契約

主スレッドが agent 別に依拠できる形式:

genshijin-investigator

<Header>:
- path:line — `symbol` — short note
集計: <counts>。

or No match. 必ずファイルパス先頭、行番号付、シンボルはバッククォート。path:\d+ で grep可能。

genshijin-builder

<path:line-range> — <change ≤10語>。
verified: <re-read OK | mismatch @ path:line>。

or 以下のいずれか: too-big. / needs-confirm. / ambiguous. / regressed. (terminal first token)。

genshijin-reviewer

path:line: <emoji> <severity>: <問題>. <修正>.
totals: N🔴 N🟡 N🔵 N❓

or No issues. ファイル → 行昇順。

チェイニングパターン

位置特定 → 修正 → 検証 (最頻):

  1. genshijin-investigator で site list 取得
  2. 主スレッドが1-2 site選び genshijin-builder にパス渡す
  3. genshijin-reviewer が diff 監査

並列スカウト (調査が広い時): 1メッセージで genshijin-investigator 2-3個並列起動 (異なる角度: defs vs callers vs tests)。主スレッドで集約。

単発編集 (sit既知時): investigator スキップ。genshijin-builder に直接 path:line 渡す。

禁止事項

  • ファイル未特定で genshijin-builder 使用禁止。先に investigator 起動 → でないと主スレッドがコンテキスト渡しでトークン消費。
  • 5ファイル refactor で genshijin-investigator → genshijin-builder チェーン禁止。Builder は too-big. 返却 → ターン浪費。
  • genshijin-reviewer に「全般フィードバック」依頼禁止 → findings のみ返却、アーキテクチャ意見なし。それ用は Code Reviewer。
  • 散文期待禁止。genshijin-crew 出力は構造化、時に cryptic。人間が直読する場合は主スレッドが言換え。

自動解除 (継承)

Subagent はセキュリティ警告・取消不可操作の確認・fragment 曖昧で誤読リスクある出力で原始人 → 通常日本語に切替。該当部分後復帰。