Focus695 · Focus695 · Sep 14, 2026 · Sep 14, 2026 · Sep 14, 2026
diff --git a/ADOPT.md b/ADOPT.md
@@ -4,7 +4,7 @@
 
 ## What you are installing
 
-BoCode gives a project two half-brains: `code/` (the source) and `book/` (the documentation stream and knowledge base), driving each other in both directions. You will install: the `book/` skeleton, the index tooling, the git policy tooling (optional), the three skills, and an `AGENTS.md` entry point. You will migrate the project's existing docs into the book. You will **not** refactor any code.
+BoCode gives a project two half-brains: `code/` (the source) and `book/` (the documentation stream and knowledge base), driving each other in both directions. You will install: the `book/` skeleton, the index tooling, the git policy tooling (optional), the four skills, and an `AGENTS.md` entry point. You will migrate the project's existing docs into the book. You will **not** refactor any code.
 
 Before starting, read in the BoCode repository: `AGENTS.md`, `book/guidelines/workflow.md`, and `skills/README.md` — this playbook assumes you know the target shape.
 
@@ -53,7 +53,7 @@ Every migrated file gets a frontmatter `description` — one plain-language line
 
 ## Step 5 — Install the skills
 
-Copy `skills/bocode`, `skills/feature-flow`, `skills/book-writeback` into the agent's skills directory (see BoCode's `skills/README.md` for per-tool locations). If the user's agent can't install skills, the workflow still works — the guidelines carry the same rules in prose.
+Copy `skills/bocode`, `skills/feature-flow`, `skills/book-writeback`, and `skills/bowrite` into the agent's skills directory (see BoCode's `skills/README.md` for per-tool locations). If the user's agent can't install skills, the workflow still works — the guidelines carry the same rules in prose.
 
 ## Step 6 — Write the project's `AGENTS.md`
 

diff --git a/AGENTS.md b/AGENTS.md
@@ -6,7 +6,7 @@ This file is the entry point for AI agents working on this repository. Humans: s
 
 <!-- Replace this paragraph in your project: what it is, who it's for, in two sentences. -->
 
-This project runs on the **BoCode workflow**: `code/` holds the source, `book/` holds the documentation stream and knowledge base. The highest architecture guidance lives inside the repository: long-term boundaries and hard constraints in `book/guidelines/architecture.md`, implementation snapshots in `book/docs/architecture/`. Internal documents never defer to external docs for authority.
+This project runs on the **BoCode workflow**: `code/` holds the source, `book/` holds the documentation stream and knowledge base. The highest architecture guidance lives inside the repository: long-term boundaries and hard constraints in `book/guidelines/architecture.md`, implementation snapshots in `book/docs/architecture/`.
 
 ## Repository layout
 

diff --git a/README.ja.md b/README.ja.md
@@ -14,7 +14,7 @@
 
 **ドキュメントがコードを駆動し、コードがドキュメントに書き戻す、プロジェクトテンプレート。**
 
-BoCode はリポジトリに二つの半脳を与えます——ソースを置く `code/` と、ドキュメントの流れとナレッジベースを置く `book/` です。両者は双方向に接続されます。今のソフトウェアが実際に書かれている方法、つまり人間と AI コーディングエージェントの共同開発を前提に設計されています。エージェントは速いが状態を持たない。プロジェクトが複利的に成長するか、セッションごとにゼロリセットされるかは、知識が耐久性のある場所に置かれているかどうかで決まります。
+BoCode はリポジトリに二つの半脳を与えます——ソースを置く `code/` と、ドキュメントの流れとナレッジベースを置く `book/` です。両者は双方向に接続されます。今のソフトウェアが実際に書かれている方法、つまり人間と AI コーディングエージェントの共同開発を前提に設計されています。プロジェクトが複利的に成長するか、セッションごとにゼロリセットされるかは、知識が耐久性のある場所に置かれているかどうかで決まります。
 
 ## 一行で導入
 
@@ -34,8 +34,8 @@ https://github.com/Focus695/BoCode を読み（まず ADOPT.md から）、そ
 
 AI エージェントで開発するプロジェクトは、ほぼ必ず三つの問題にぶつかります。
 
-1. **エージェントは状態を持たない。** セッションは毎回ゼロから始まります。書き残されていない知識は、再導出され、再質問され、再破壊されます——毎回定価を払いながら。
-2. **ドキュメントは腐る。** 立ち上げ時に一度書かれ、二度と更新されない。更新の利益は後で、スキップの利益は今なので、常に負けます。やがて誰も信じなくなり、当然です。
+1. **エージェントは状態を持たない。** セッションは毎回ゼロから始まります。書き残されていない知識は、再導出され、再質問され、再破壊されます。
+2. **ドキュメントは腐る。** 立ち上げ時に一度書かれ、二度と更新されない。更新の利益は後で、スキップの利益は今なので、更新は常に後回しにされます。やがて誰も信じなくなります——古いドキュメントは信じる価値がありません。
 3. **人間がハンドルを失う。** エージェントは頼まれたものに加えて、頼まれていない十二個の変更も実装します。ゲートのないスピードは、スコープの漂移と意思決定の未記録と、「なぜ」の蒸発を意味します。
 
 三つとも同じ根っこを指しています：知識と制御は、チャット履歴でも誰かの記憶でもない場所に住む必要がある、と。
@@ -117,7 +117,7 @@ diff を読まないと分からない summary は summary ではありません
 
 ### インデックス契約
 
-ナビゲートできないナレッジベースは、書き込み専用のディスクです。BoCode は発見可能性をビルドチェックにします：
+見つけられないドキュメントは、存在しないのと同じです。BoCode は発見可能性をビルドチェックにします：
 
 - すべての book ドキュメントは frontmatter の `description` で始まる——体裁ではなく内容を語る、人の言葉で一行
 - `code/tools/gen-book-index.mjs`（依存ゼロ、プレーン Node）が `book/README.md`——全ドキュメントのマスターインデックス——を再生成する
@@ -127,7 +127,7 @@ diff を読まないと分からない summary は summary ではありません
 
 ### 人の言葉で書く
 
-ドキュメントは何年も読まれ、数分で書かれる。だからスタイル規則は短く厳格です（`book/guidelines/writing-style.md`）：具体的な人が具体的な状況で話すように書く——プロフェッショナルで構わないが、テンプレート的なのは不可。埋め草の前置きと空っぽのまとめを削り、事実は固定する——数値・コマンド・名前・責任の所在は動かさない。中国語の完全なルールセットは [shuorenhua](https://github.com/MrGeDiao/shuorenhua) スキルに、guideline ファイルは言語非依存のデフォルトです。
+ドキュメントは何年も読まれ、数分で書かれる。だからスタイル規則は短く厳格です（`book/guidelines/writing-style.md`）：具体的な人が具体的な状況で話すように書く——プロフェッショナルで構わないが、テンプレート的なのは不可。埋め草の前置きと空っぽのまとめを削り、事実は固定する——数値・コマンド・名前・責任の所在は動かさない。この基準は `bowrite` スキル（薄写：薄く書く——文は減り、核は変わらない；うまく書く——自然に、直接に）がエージェント側で担い、MrGeDiao の [shuorenhua](https://github.com/MrGeDiao/shuorenhua) と本プロジェクトのレビューで蓄積したパターンから蒸留されています。
 
 ### クリーンな git フロー
 
@@ -149,8 +149,8 @@ git clone <your-fork-url> myproject && cd myproject
 # 2. テンプレートをあなたのプロジェクトに変える（名前、説明、記録のクリーンアップ）：
 bash scripts/init-project.sh myproject "What it does, in one line"
 
-# 3. AI エージェントに三つのスキルをインストール：
-cp -r skills/bocode skills/feature-flow skills/book-writeback <your-skills-dir>/
+# 3. AI エージェントに四つのスキルをインストール：
+cp -r skills/bocode skills/feature-flow skills/book-writeback skills/bowrite <your-skills-dir>/
 #    (ZCode: ~/.zcode/skills/ · Claude Code: ~/.claude/skills/ — 詳細は skills/README.md)
 
 # 4. エージェントを AGENTS.md に向ける——ほとんどのツールが自動で読みます。
@@ -169,7 +169,7 @@ cp -r skills/bocode skills/feature-flow skills/book-writeback <your-skills-dir>/
 | `book/plans/` | 実装プラン（施工図） |
 | `book/notes/{summary,learn,task,issue}/` | 四種類のノート |
 | `book/docs/{architecture,api,decisions}/` | スナップショット、契約、ADR |
-| `skills/` | `bocode`、`feature-flow`、`book-writeback` |
+| `skills/` | `bocode`、`feature-flow`、`book-writeback`、`bowrite` |
 | `code/tools/gen-book-index.mjs` | インデックスジェネレータ |
 | `scripts/init-project.sh` | テンプレート → あなたのプロジェクト |
 | `scripts/check-commit-message.mjs` | Clean Commit バリデータ（`.githooks/` 付き） |
@@ -193,7 +193,7 @@ cp -r skills/bocode skills/feature-flow skills/book-writeback <your-skills-dir>/
 
 ## クレジット
 
-BoCode の git 規律は WGTech Labs の二つのオープンスタンダード——[Clean Commit](https://github.com/wgtechlabs/clean-commit)（コミットメッセージ形式）と [Clean Flow](https://github.com/wgtechlabs/clean-flow)（ブランチモデル）——の上に構築され、両スペックが記述するが同梱しない実行ツールを追加しています。人の言葉で書くスタイルは MrGeDiao の [shuorenhua](https://github.com/MrGeDiao/shuorenhua) に由来します。
+BoCode の git 規律は WGTech Labs の二つのオープンスタンダード——[Clean Commit](https://github.com/wgtechlabs/clean-commit)（コミットメッセージ形式）と [Clean Flow](https://github.com/wgtechlabs/clean-flow)（ブランチモデル）——の上に構築され、両スペックが記述するが同梱しない実行ツールを追加しています。人の言葉で書くスタイルは MrGeDiao の [shuorenhua](https://github.com/MrGeDiao/shuorenhua) に由来し、`bowrite` スキルへ蒸留されています。
 
 ## ライセンス
 

diff --git a/README.md b/README.md
@@ -14,7 +14,7 @@
 
 **A project template where documentation drives code, and code writes back.**
 
-BoCode gives a repository two half-brains — `code/` for the source, `book/` for the documentation stream and knowledge base — wired together in both directions. It is built for the way software is actually written now: by a human and AI coding agents together, where agents are fast but stateless, and the difference between a project that compounds and one that resets every session is whether its knowledge lives somewhere durable.
+BoCode gives a repository two half-brains — `code/` for the source, `book/` for the documentation stream and knowledge base — wired together in both directions. It is built for the way software is written now: a human and AI coding agents together. Whether a project compounds or resets every session comes down to whether its knowledge lives somewhere durable.
 
 ## Adopt it in one line
 
@@ -26,7 +26,7 @@ Read https://github.com/Focus695/BoCode — start with ADOPT.md — and retrofit
 
 Starting from scratch? The three-command Quick Start is below.
 
-Clone it, run one script, install three skills, and your project gains: a gated development workflow, a self-indexing knowledge base, a write-back discipline that turns every feature into accumulated experience, and a git policy that keeps history readable.
+Clone it, run one script, install four skills, and your project gains: a gated development workflow, a self-indexing knowledge base, a write-back discipline that turns every feature into accumulated experience, and a git policy that keeps history readable.
 
 ---
 
@@ -36,8 +36,8 @@ Clone it, run one script, install three skills, and your project gains: a gated
 
 Three failures show up in almost every project built with AI agents:
 
-1. **Agents are stateless.** Every session starts from zero. Knowledge that isn't written down somewhere durable gets re-derived, re-asked, and re-broken — at full price, every time.
-2. **Documentation rots.** Written once at the start, updated never. Because updating it pays off later and skipping it pays off now, it always loses. Eventually nobody trusts it, and reasonably so.
+1. **Agents are stateless.** Every session starts from zero. Knowledge that isn't written down somewhere durable gets re-derived, re-asked, and re-broken.
+2. **Documentation rots.** Written once at the start, updated never. Because updating it pays off later and skipping it pays off now, it always gets skipped. Eventually nobody trusts it — a stale document doesn't deserve trust.
 3. **Humans lose the wheel.** An agent will implement what you asked plus twelve things you didn't. Speed without gates means scope drifts, decisions go unrecorded, and the project's "why" evaporates.
 
 Each failure points at the same root: knowledge and control need a place to live that isn't chat history or someone's memory.
@@ -102,7 +102,7 @@ Every feature runs a six-phase workflow with hard gates (`book/guidelines/workfl
 | 5 — Review | findings list | Recorded, never fixed |
 | 6 — Wrap-up | summary, learn, issues, changelog, index | Recorded, never fixed |
 
-The design intent: agents compress implementation to minutes, which moves the bottleneck to decisions and scope. The gates put human hands exactly there — Step 0 fills in what the request didn't say, Phase 2 is where a person approves direction, and Phases 5–6 enforce the separation of observing and fixing. **A problem found at review becomes a recorded issue; fixing it is separate work.** The gate that didn't let you fix it in passing is the same gate that keeps the review honest.
+The design intent: agents compress implementation to minutes, which moves the bottleneck to decisions and scope. The gates hold exactly there — Step 0 fills in what the request didn't say, Phase 2 is where a person approves direction, and Phases 5–6 enforce the separation of observing and fixing. **A problem found at review becomes a recorded issue; fixing it is separate work.** That is what keeps the review honest.
 
 ### Write for the reader
 
@@ -119,7 +119,7 @@ A summary that requires reading the diff isn't a summary. A learn entry whose ti
 
 ### The index contract
 
-A knowledge base you can't navigate is a write-only disk. BoCode makes discoverability a build check:
+Documentation you can't find doesn't exist. BoCode makes discoverability a build check:
 
 - Every book document starts with a frontmatter `description` — one plain-language line about the *content*, not the genre.
 - `code/tools/gen-book-index.mjs` (zero dependencies, plain Node) regenerates `book/README.md` — the master index of every document, grouped by directory.
@@ -129,7 +129,7 @@ A knowledge base you can't navigate is a write-only disk. BoCode makes discovera
 
 ### Speak human
 
-Documentation is read for years and written in minutes, so the style rule is strict and short (`book/guidelines/writing-style.md`): write like a specific person in a specific situation — professional is fine, templated is not. Cut filler openers and empty summaries; keep the facts locked — numbers, commands, names, and responsibility attribution never move. For Chinese writing, the full ruleset is the [shuorenhua](https://github.com/MrGeDiao/shuorenhua) skill; the guideline file is the language-agnostic default.
+Documentation is read for years and written in minutes, so the style rule is strict and short (`book/guidelines/writing-style.md`): write like a specific person in a specific situation — professional is fine, templated is not. Cut filler openers and empty summaries; keep the facts locked — numbers, commands, names, and responsibility attribution never move. The `bowrite` skill（薄写 — write thin, write well: fewer words, same core）carries this standard for agents; it distills [shuorenhua](https://github.com/MrGeDiao/shuorenhua) by MrGeDiao together with patterns accumulated in this project's own reviews.
 
 ### Clean git flow
 
@@ -151,8 +151,8 @@ git clone <your-fork-url> myproject && cd myproject
 # 2. Turn the template into your project (name, description, clean records):
 bash scripts/init-project.sh myproject "What it does, in one line"
 
-# 3. Install the three skills for your AI agent:
-cp -r skills/bocode skills/feature-flow skills/book-writeback <your-skills-dir>/
+# 3. Install the four skills for your AI agent:
+cp -r skills/bocode skills/feature-flow skills/book-writeback skills/bowrite <your-skills-dir>/
 #    (ZCode: ~/.zcode/skills/ · Claude Code: ~/.claude/skills/ — see skills/README.md)
 
 # 4. Point your agent at AGENTS.md — most tools read it automatically.
@@ -171,7 +171,7 @@ Requirements: `bash` and `node` for the two scripts. No packages, no install ste
 | `book/plans/` | Implementation blueprints |
 | `book/notes/{summary,learn,task,issue}/` | The four note types |
 | `book/docs/{architecture,api,decisions}/` | Snapshots, contracts, ADRs |
-| `skills/` | `bocode`, `feature-flow`, `book-writeback` |
+| `skills/` | `bocode`, `feature-flow`, `book-writeback`, `bowrite` |
 | `code/tools/gen-book-index.mjs` | The index generator |
 | `scripts/init-project.sh` | Template → your project |
 | `scripts/check-commit-message.mjs` | Clean Commit validator (+ `.githooks/`) |
@@ -195,7 +195,7 @@ Months are stable, low-maintenance physical grouping. Features are found through
 
 ## Credits
 
-BoCode's git discipline builds on two open standards by WGTech Labs — [Clean Commit](https://github.com/wgtechlabs/clean-commit) (commit message format) and [Clean Flow](https://github.com/wgtechlabs/clean-flow) (branching model) — and adds the enforcement tooling both specs describe but don't ship. The plain-language writing standard draws on [shuorenhua](https://github.com/MrGeDiao/shuorenhua) by MrGeDiao.
+BoCode's git discipline builds on two open standards by WGTech Labs — [Clean Commit](https://github.com/wgtechlabs/clean-commit) (commit message format) and [Clean Flow](https://github.com/wgtechlabs/clean-flow) (branching model) — and adds the enforcement tooling both specs describe but don't ship. The plain-language writing standard draws on [shuorenhua](https://github.com/MrGeDiao/shuorenhua) by MrGeDiao, distilled into the `bowrite` skill.
 
 ## License