📦 deps(thirdparty): update snapshots

2026-08-24 09:30:03 +08:00
parent 14e7209e78
commit 5d88510274
153 changed files with 3428 additions and 1203 deletions
@@ -9,7 +9,7 @@
    {
      "name": "brooks-lint",
      "description": "AI code reviews grounded in twelve classic engineering books — decay risk diagnostics with book citations, severity labels, and six analysis modes (PR review, architecture audit, tech debt, test quality, health dashboard, full-sweep auto-fix)",
-      "version": "1.4.3",
+      "version": "1.5.0",
      "source": "./",
      "author": {
        "name": "hyhmrright",
@@ -1,7 +1,7 @@
 {
  "name": "brooks-lint",
  "description": "AI code reviews grounded in twelve classic engineering books — decay risk diagnostics with book citations, severity labels, and six analysis modes (PR review, architecture audit, tech debt, test quality, health dashboard, full-sweep auto-fix)",
-  "version": "1.4.3",
+  "version": "1.5.0",
  "author": {
    "name": "hyhmrright",
    "email": "hyhmrright@gmail.com"
@@ -1,6 +1,6 @@
 {
  "name": "brooks-lint",
-  "version": "1.4.3",
+  "version": "1.5.0",
  "description": "AI code reviews grounded in twelve classic engineering books — decay risk diagnostics with book citations, severity labels, and six analysis modes (PR review, architecture audit, tech debt, test quality, health dashboard, full-sweep auto-fix)",
  "author": {
    "name": "hyhmrright",
@@ -24,6 +24,9 @@ inputs:
  model:
    description: "Claude model to use"
    default: "claude-sonnet-4-6"
+  api-base-url:
+    description: "Anthropic-compatible /v1/messages endpoint to call instead of api.anthropic.com (empty = Anthropic)"
+    default: ""

 outputs:
  score:
@@ -71,10 +74,17 @@ runs:
      shell: bash
      env:
        ANTHROPIC_API_KEY: ${{ inputs.anthropic-api-key }}
+        API_BASE_URL: ${{ inputs.api-base-url }}
        SARIF_FILE: ${{ inputs.sarif-file }}
        MODE: ${{ inputs.mode }}
        MODEL: ${{ inputs.model }}
      run: |
+        # Exported only when the input is set: naming ANTHROPIC_BASE_URL in the
+        # env block above would blank out one inherited from the job env. The
+        # SDK reads the variable itself, so nothing downstream needs a flag.
+        if [ -n "$API_BASE_URL" ]; then
+          export ANTHROPIC_BASE_URL="$API_BASE_URL"
+        fi
        sarif_arg=()
        if [ -n "$SARIF_FILE" ]; then
          sarif_arg=(--sarif-out "$SARIF_FILE")
@@ -0,0 +1,40 @@
+name: Star history
+
+# GitHub restricted the stargazers API to a repo's own admins and collaborators,
+# so no third-party service can chart this repo any more. GITHUB_TOKEN can read
+# it, so we regenerate the committed SVG here instead.
+on:
+  schedule:
+    - cron: "17 3 * * 1" # Mondays, 03:17 UTC
+  workflow_dispatch:
+
+concurrency:
+  group: star-history
+  cancel-in-progress: true
+
+jobs:
+  refresh:
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    permissions:
+      contents: write
+    steps:
+      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
+      - name: Set up Node.js
+        uses: actions/setup-node@60edb5dd545a775178f52524783378180af0d1f8 # v4
+        with:
+          node-version: 22
+      - name: Regenerate the chart
+        run: node scripts/gen-star-history.mjs
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+      - name: Commit when the chart moved
+        run: |
+          if git diff --quiet -- assets/star-history.svg assets/star-history.json; then
+            echo "No change."
+            exit 0
+          fi
+          git config user.name "github-actions[bot]"
+          git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
+          git commit -m "chore: refresh the star history chart" -- assets/star-history.svg assets/star-history.json
+          git push
@@ -2,6 +2,28 @@

 All notable changes to brooks-lint are documented here.

+## [1.5.0] - 2026-08-14
+
+### Added
+
+- **DeepSeek Harness (`dsh`) support** — DeepSeek AI's open-source agent harness
+  loads standard Agent Skills through `packages/skill/skill-filesystem`, so all
+  six modes run with no conversion. `./scripts/install.sh dsh` installs into
+  `$DSH_HOME/skills` (default `~/.dsh/skills`); `--project` targets
+  `./.dsh/skills`. dsh also scans `~/.agents/skills`, so the existing
+  vendor-neutral `install.sh agents` install already covered it. New guide at
+  [`docs/dsh-setup.md`](docs/dsh-setup.md), with the platform added to the README
+  install table in all six languages and to `docs/getting-started.md`.
+
+  Three details of dsh's loader matter and are documented in the guide: discovery
+  is one level deep (`<root>/<name>/SKILL.md`), which the installer's flat layout
+  already satisfies so `../_shared/` resolves; skill names must be kebab-case,
+  which `brooks-*` already is; and dsh recognises a whitespace-bounded `/name`
+  token anywhere in a message, so `/brooks-review` and the other five work from
+  its `/` menu or typed inline. dsh also reads `AGENTS.md` — `$DSH_HOME/AGENTS.md`
+  plus every file from the project root down to the working directory — so the
+  repo's Iron Law and Health Score rules load the same way they do elsewhere.
+
 ## [1.4.3] - 2026-08-04

 ### Fixed
@@ -36,6 +36,8 @@ Guidance for Claude Code when modifying this repository. For repo layout, instal
  - `` field `commands` is not accepted `` — `"commands": []` is what stops the CLI from migrating `commands/*.md` into duplicate `source-command-brooks-*` skills (issue #22). Confirmed by installing both ways under an isolated `CODEX_HOME`: drop the field and Codex regenerates `.codex-plugin/migrated-command-skills/`. The CLI's `RawPluginManifest` accepts `commands`; only the ingestion schema rejects it.
  - `` skill `_shared` is missing `SKILL.md` `` — the validator only skips dot-prefixed dirs, and `_shared/` is a shared-framework dir by design (see above). Renaming it to `.shared/` would also require changing `install.sh`'s `cp -R "$SRC"/*`, whose glob does not match dot-dirs.
 - **GitHub Action cache:** `.github/actions/brooks-lint/action.yml` uses `actions/cache@v4` with built-in cache-hit guard — do NOT add a manual directory check.
+- **Adding a platform is derived, not hardcoded:** create `docs/<name>-setup.md`, add the platform to `install.sh`'s `PLATFORMS` **and** both `global_dir()` / `project_dir()` case tables, then link the new guide from every `README*.md` and `docs/getting-started.md`. `npm run validate` derives both sides (`scripts/platforms.mjs`) and fails on any gap — a guide missing from one translation, or a `PLATFORMS` entry with no directory mapping. Do not add a hand-maintained platform list anywhere.
+- **Star history is data-first, never hand-drawn:** `assets/star-history.json` (raw `starred_at` timestamps, no usernames) is the source of truth; `assets/star-history.svg` is a pure function of it. Never hand-edit the SVG — `npm run validate` re-renders and fails on any mismatch. `node scripts/gen-star-history.mjs` refetches (needs `GITHUB_TOKEN` or `gh auth`, since GitHub restricted the stargazers API to admins/collaborators on 2026-06-30); `--render-only` redraws offline from the committed data. The render must stay deterministic — anchoring the time axis to the clock would make the weekly workflow commit noise on every run.
 - **Custom risks:** Teams add project-specific risk codes via `custom-risks-guide.md` in their project root. Template lives at `skills/_shared/custom-risks-guide.md`.

 ## How the Skills Work
@@ -27,7 +27,7 @@
 </p>

 <p align="center">
-  <img src="https://img.shields.io/badge/version-1.4.3-blue.svg" alt="Version">
+  <img src="https://img.shields.io/badge/version-1.5.0-blue.svg" alt="Version">
  <img src="https://img.shields.io/badge/license-MIT-green.svg" alt="MIT License">
  <img src="https://img.shields.io/badge/Claude_Code-Plugin-blueviolet.svg" alt="Claude Code Plugin">
  <img src="https://img.shields.io/badge/Codex_CLI-Skill-orange.svg" alt="Codex CLI Skill">
@@ -78,7 +78,7 @@ Luego solo pide ("revisa este PR", "audita la arquitectura"), o ejecuta uno de l
 ([qué hace cada uno](#comandos-de-barra)).

 Cada hallazgo se devuelve como **Síntoma → Origen → Consecuencia → Remedio** con una cita de libro y
-una puntuación de salud de 0 a 100. Las opciones completas de instalación (8 plataformas más) y la
+una puntuación de salud de 0 a 100. Las opciones completas de instalación (9 plataformas más) y la
 configuración de CI/CD están [más abajo](#instalación).

 ## Los doce libros
@@ -273,7 +273,7 @@ Install the brooks-lint skill from hyhmrright/brooks-lint       # pídelo dentro

 O usa el instalador de abajo: `./scripts/install.sh gemini` / `./scripts/install.sh codex`.

-### Cualquier otra plataforma — OpenCode · Cursor · Windsurf · Antigravity · pi · Copilot · Kiro · Factory Droid
+### Cualquier otra plataforma — OpenCode · Cursor · Windsurf · Antigravity · pi · Copilot · Kiro · Factory Droid · DeepSeek Harness

 brooks-lint se distribuye como [Agent Skills](https://agentskills.io) estándar. **Cualquier agente que cargue Agent
 Skills ejecuta los seis modos sin conversión alguna** — un solo comando los instala:
@@ -281,7 +281,7 @@ Skills ejecuta los seis modos sin conversión alguna** — un solo comando los i
 ```bash
 # elige tu plataforma; --project instala en el repositorio actual en lugar de en tu configuración global
 curl -fsSL https://raw.githubusercontent.com/hyhmrright/brooks-lint/main/scripts/install.sh | bash -s -- <platform>
-#   <platform> = opencode · cursor · windsurf · antigravity · pi · kiro · copilot · droid · gemini · codex · agents
+#   <platform> = opencode · cursor · windsurf · antigravity · pi · kiro · copilot · droid · dsh · gemini · codex · agents
 ```

 El instalador copia los skills **de forma plana** en la carpeta correcta, de modo que el framework compartido
@@ -298,11 +298,12 @@ El instalador copia los skills **de forma plana** en la carpeta correcta, de mod
 | GitHub Copilot | `.github/skills` (`--project`) | `.claude/skills`, `AGENTS.md` | [configuración](docs/copilot-setup.md) |
 | Kiro (AWS) | `~/.kiro/skills` | `AGENTS.md` | [configuración](docs/kiro-setup.md) |
 | Factory Droid | `~/.factory/skills` | `AGENTS.md` | [configuración](docs/factory-droid-setup.md) |
+| DeepSeek Harness (`dsh`) | `~/.dsh/skills` | `~/.agents/skills`, `AGENTS.md` | [configuración](docs/dsh-setup.md) |

-Kiro y Factory Droid también registran `/brooks-review` automáticamente. ¿Nuevo en los skills, o usas un
-agente que no aparece aquí? Consulta **[docs/getting-started.md](docs/getting-started.md)**.
+Kiro, Factory Droid y DeepSeek Harness también registran `/brooks-review` automáticamente. ¿Nuevo en los
+skills, o usas un agente que no aparece aquí? Consulta **[docs/getting-started.md](docs/getting-started.md)**.

-> **🧪 Estado de verificación.** Claude Code, Gemini CLI y Codex CLI están verificados por el mantenedor. Las ocho
+> **🧪 Estado de verificación.** Claude Code, Gemini CLI y Codex CLI están verificados por el mantenedor. Las nueve
 > plataformas anteriores están documentadas a partir de la especificación oficial de skills de cada herramienta y verificadas a nivel
 > de diseño de archivos (el instalador está probado), pero el mantenedor aún no las ha ejecutado de extremo a extremo en cada plataforma. ¿Probaste
 > alguna — funciona **o** está rota? [Abre un issue](https://github.com/hyhmrright/brooks-lint/issues/new) con
@@ -322,10 +323,11 @@ agente que no aparece aquí? Consulta **[docs/getting-started.md](docs/getting-s

 **Sintaxis por plataforma.** Claude Code también acepta la forma con espacio de nombres
 `/brooks-lint:brooks-review` — las formas cortas las instala el hook session-start al iniciar la primera
-sesión. Codex CLI usa `$brooks-review`. Gemini CLI usa la tabla tal cual. OpenCode, Cursor, Antigravity y pi
-invocan los Agent Skills desde la `description` de cada skill, así que basta con pedirlo ("revisa este PR",
-"¿dónde está nuestra peor deuda técnica?"); para invocarlos explícitamente usa la sintaxis propia de cada
-plataforma (pi registra cada skill como `/skill:brooks-review`). En todas las plataformas los skills también
+sesión. Codex CLI usa `$brooks-review`. Gemini CLI usa la tabla tal cual. OpenCode, Cursor, Antigravity, pi y
+DeepSeek Harness invocan los Agent Skills desde la `description` de cada skill, así que basta con pedirlo
+("revisa este PR", "¿dónde está nuestra peor deuda técnica?"); para invocarlos explícitamente usa la sintaxis
+propia de cada plataforma (pi registra cada skill como `/skill:brooks-review`; dsh usa la tabla tal cual, desde
+su menú `/` o escrito a mano). En todas las plataformas los skills también
 se activan solos cuando hablas de calidad de código, arquitectura o salud de las pruebas.

 > Las revisiones de PR incluyen automáticamente una comprobación rápida de pruebas (Step 7, ligera; se omite
@@ -448,6 +450,18 @@ La action publica la revisión como un comentario del PR y, opcionalmente, hace

 `fail-on-regression` lee `.brooks-lint-history.json`, así que confirma ese archivo para imponer "sin nuevas regresiones". Definir `sarif-file` hace que los hallazgos aparezcan en línea en la pestaña **Files changed** del PR y requiere el permiso `security-events: write` en el job.

+**Endpoint de API personalizado.** `api-base-url` apunta la action a cualquier endpoint `/v1/messages` compatible con Anthropic —un proxy autoalojado, una pasarela de LLM, un espejo regional— en lugar de `api.anthropic.com`. Pasa la clave de ese endpoint como `anthropic-api-key` y el id de modelo que espera como `model`:
+
+```yaml
+        with:
+          mode: review
+          api-base-url: https://your-gateway.example.com
+          anthropic-api-key: ${{ secrets.GATEWAY_API_KEY }}
+          model: gateway-model-id
+```
+
+brooks-lint envía tu diff al host que indiques aquí, así que apúntalo solo a uno en el que confíes con tu código fuente. Ejecutar `scripts/ci-review.mjs` por tu cuenta no necesita ninguna opción: el SDK de Anthropic lee `ANTHROPIC_BASE_URL` directamente.
+
 **Coste:** ~$0,05–0,15 por ejecución de PR, según el tamaño del diff y el modelo. Se recomienda ejecutar solo en eventos `pull_request`.

 ## Hoja de ruta
@@ -486,7 +500,7 @@ síntesis de sus ideas, aplicada a la evaluación moderna de la calidad del cód

 ## Historial de estrellas

-[![Star History Chart](https://api.star-history.com/svg?repos=hyhmrright/brooks-lint&type=Date)](https://star-history.com/#hyhmrright/brooks-lint&Date)
+[![Star History](assets/star-history.svg)](https://github.com/hyhmrright/brooks-lint/stargazers)

 ---

@@ -27,7 +27,7 @@
 </p>

 <p align="center">
-  <img src="https://img.shields.io/badge/version-1.4.3-blue.svg" alt="Version">
+  <img src="https://img.shields.io/badge/version-1.5.0-blue.svg" alt="Version">
  <img src="https://img.shields.io/badge/license-MIT-green.svg" alt="MIT License">
  <img src="https://img.shields.io/badge/Claude_Code-Plugin-blueviolet.svg" alt="Claude Code Plugin">
  <img src="https://img.shields.io/badge/Codex_CLI-Skill-orange.svg" alt="Codex CLI Skill">
@@ -78,7 +78,7 @@ curl -fsSL https://raw.githubusercontent.com/hyhmrright/brooks-lint/main/scripts
 `/brooks-health`、`/brooks-sweep`（[それぞれの機能](#スラッシュコマンド)）。

 すべての指摘は **症状 → 根源 → 結果 → 対策** の形式で、書籍の出典と 0〜100 の健全性スコアとともに
-返されます。完全なインストール方法（さらに 8 つのプラットフォーム）と CI/CD のセットアップは
+返されます。完全なインストール方法（さらに 9 つのプラットフォーム）と CI/CD のセットアップは
 [以下](#インストール)を参照してください。

 ## 十二冊の書籍
@@ -273,7 +273,7 @@ Install the brooks-lint skill from hyhmrright/brooks-lint       # Codex セッ

 または下記のインストーラーを使用：`./scripts/install.sh gemini` / `./scripts/install.sh codex`。

-### その他すべてのプラットフォーム — OpenCode · Cursor · Windsurf · Antigravity · pi · Copilot · Kiro · Factory Droid
+### その他すべてのプラットフォーム — OpenCode · Cursor · Windsurf · Antigravity · pi · Copilot · Kiro · Factory Droid · DeepSeek Harness

 brooks-lint は標準的な [Agent Skills](https://agentskills.io) として配布されています。**Agent
 Skills を読み込むエージェントなら、どれも変換なしで六つすべてのモードを実行できます**——1 つのコマンドでインストールできます：
@@ -281,7 +281,7 @@ Skills を読み込むエージェントなら、どれも変換なしで六つ
 ```bash
 # プラットフォームを選択；--project はグローバル設定ではなく現在のリポジトリにインストール
 curl -fsSL https://raw.githubusercontent.com/hyhmrright/brooks-lint/main/scripts/install.sh | bash -s -- <platform>
-#   <platform> = opencode · cursor · windsurf · antigravity · pi · kiro · copilot · droid · gemini · codex · agents
+#   <platform> = opencode · cursor · windsurf · antigravity · pi · kiro · copilot · droid · dsh · gemini · codex · agents
 ```

 インストーラーはスキルをあなたのプラットフォームに適したフォルダへ**フラット**にコピーするため、共有フレームワーク
@@ -299,11 +299,12 @@ curl -fsSL https://raw.githubusercontent.com/hyhmrright/brooks-lint/main/scripts
 | GitHub Copilot | `.github/skills`（`--project`） | `.claude/skills`、`AGENTS.md` | [設定](docs/copilot-setup.md) |
 | Kiro（AWS） | `~/.kiro/skills` | `AGENTS.md` | [設定](docs/kiro-setup.md) |
 | Factory Droid | `~/.factory/skills` | `AGENTS.md` | [設定](docs/factory-droid-setup.md) |
+| DeepSeek Harness（`dsh`） | `~/.dsh/skills` | `~/.agents/skills`、`AGENTS.md` | [設定](docs/dsh-setup.md) |

-Kiro と Factory Droid は `/brooks-review` も自動登録します。スキルが初めて、または上記にないエージェントを
-お使いですか？ **[docs/getting-started.md](docs/getting-started.md)** を参照してください。
+Kiro、Factory Droid、DeepSeek Harness は `/brooks-review` も自動登録します。スキルが初めて、または
+上記にないエージェントをお使いですか？ **[docs/getting-started.md](docs/getting-started.md)** を参照してください。

-> **🧪 検証状況。** Claude Code、Gemini CLI、Codex CLI はメンテナーによって検証済みです。上記の八つの
+> **🧪 検証状況。** Claude Code、Gemini CLI、Codex CLI はメンテナーによって検証済みです。上記の九つの
 > プラットフォームは各ツールの公式スキル仕様から文書化され、ファイルレイアウトのレベルで検証されています
 > （インストーラーはテスト済み）が、メンテナーがすべてのプラットフォームでエンドツーエンドに実行したわけ
 > ではまだありません。どれかを試した——動いた **または** 壊れた？ プラットフォーム、バージョン、見たこと
@@ -323,10 +324,11 @@ Kiro と Factory Droid は `/brooks-review` も自動登録します。スキル

 **プラットフォーム別の構文。** Claude Code は名前空間付きの完全形 `/brooks-lint:brooks-review` も受け付けます
 ——短縮形は session-start フックが最初のセッション開始時に自動インストールします。Codex CLI は
-`$brooks-review`。Gemini CLI は上の表のとおり。OpenCode、Cursor、Antigravity、pi は各スキルの
-`description` から Agent Skills を呼び出すので、話しかけるだけで十分です（「この PR をレビューして」
+`$brooks-review`。Gemini CLI は上の表のとおり。OpenCode、Cursor、Antigravity、pi、DeepSeek Harness は
+各スキルの `description` から Agent Skills を呼び出すので、話しかけるだけで十分です（「この PR をレビューして」
 「最悪の技術的負債はどこ？」）。明示的に呼び出す場合は各プラットフォームの構文を使います（pi は各スキルを
-`/skill:brooks-review` として登録）。どのプラットフォームでも、コード品質・アーキテクチャ・テストの健全性に
+`/skill:brooks-review` として登録。dsh は上の表のとおりで、`/` メニューから選ぶか直接入力）。どの
+プラットフォームでも、コード品質・アーキテクチャ・テストの健全性に
 ついて話すと、スキルは自動的にトリガーされます。

 > PR レビューには軽量な Step 7 クイックテストチェックが自動的に含まれます（ドキュメントのみの diff では
@@ -448,6 +450,18 @@ jobs:

 `fail-on-regression` は `.brooks-lint-history.json` を読み取るため、そのファイルをコミットすれば「新たな回帰なし」を強制できます。`sarif-file` を設定すると、指摘が PR の **Files changed** タブにインラインで表示されるようになり、ジョブに `security-events: write` 権限が必要になります。

+**カスタム API エンドポイント。** `api-base-url` を指定すると、action は `api.anthropic.com` ではなく Anthropic 互換の `/v1/messages` エンドポイント（自前のプロキシ、LLM ゲートウェイ、リージョンミラーなど）を呼び出します。そのエンドポイントのキーを `anthropic-api-key` に、期待されるモデル id を `model` に渡してください：
+
+```yaml
+        with:
+          mode: review
+          api-base-url: https://your-gateway.example.com
+          anthropic-api-key: ${{ secrets.GATEWAY_API_KEY }}
+          model: gateway-model-id
+```
+
+brooks-lint はここで指定したホストに diff を送信するため、ソースコードを預けられる相手だけを指定してください。`scripts/ci-review.mjs` を自分で実行する場合はフラグは不要です — Anthropic SDK が `ANTHROPIC_BASE_URL` を直接読み取ります。
+
 **コスト：** PR 実行ごとにおよそ $0.05〜0.15、diff のサイズとモデルによります。`pull_request` イベントのみで実行することを推奨します。

 ## ロードマップ
@@ -486,7 +500,7 @@ MIT License — 詳細は [LICENSE](LICENSE) を参照してください。

 ## スター履歴

-[![Star History Chart](https://api.star-history.com/svg?repos=hyhmrright/brooks-lint&type=Date)](https://star-history.com/#hyhmrright/brooks-lint&Date)
+[![Star History](assets/star-history.svg)](https://github.com/hyhmrright/brooks-lint/stargazers)

 ---

@@ -27,7 +27,7 @@
 </p>

 <p align="center">
-  <img src="https://img.shields.io/badge/version-1.4.3-blue.svg" alt="Version">
+  <img src="https://img.shields.io/badge/version-1.5.0-blue.svg" alt="Version">
  <img src="https://img.shields.io/badge/license-MIT-green.svg" alt="MIT License">
  <img src="https://img.shields.io/badge/Claude_Code-Plugin-blueviolet.svg" alt="Claude Code Plugin">
  <img src="https://img.shields.io/badge/Codex_CLI-Skill-orange.svg" alt="Codex CLI Skill">
@@ -78,7 +78,7 @@ curl -fsSL https://raw.githubusercontent.com/hyhmrright/brooks-lint/main/scripts
 `/brooks-sweep`（[각각 하는 일](#슬래시-명령)）.

 모든 진단은 도서 출처와 0–100 건강 점수와 함께 **증상 → 근원 → 결과 → 처방** 형태로 돌아옵니다. 전체
-설치 옵션（추가 8개 플랫폼）과 CI/CD 설정은 [아래](#설치)를 참고하세요.
+설치 옵션（추가 9개 플랫폼）과 CI/CD 설정은 [아래](#설치)를 참고하세요.

 ## 열두 권의 책

@@ -272,7 +272,7 @@ Install the brooks-lint skill from hyhmrright/brooks-lint       # Codex 세션

 또는 아래 설치기를 사용하세요: `./scripts/install.sh gemini` / `./scripts/install.sh codex`.

-### 그 밖의 모든 플랫폼 — OpenCode · Cursor · Windsurf · Antigravity · pi · Copilot · Kiro · Factory Droid
+### 그 밖의 모든 플랫폼 — OpenCode · Cursor · Windsurf · Antigravity · pi · Copilot · Kiro · Factory Droid · DeepSeek Harness

 brooks-lint는 표준 [Agent Skills](https://agentskills.io) 형태로 배포됩니다. **Agent
 Skills를 로드하는 모든 에이전트는 변환 없이 여섯 가지 모드를 모두 실행합니다** — 한 줄의 명령으로 설치됩니다:
@@ -280,7 +280,7 @@ Skills를 로드하는 모든 에이전트는 변환 없이 여섯 가지 모드
 ```bash
 # 플랫폼을 고르세요; --project는 전역 설정 대신 현재 저장소에 설치합니다
 curl -fsSL https://raw.githubusercontent.com/hyhmrright/brooks-lint/main/scripts/install.sh | bash -s -- <platform>
-#   <platform> = opencode · cursor · windsurf · antigravity · pi · kiro · copilot · droid · gemini · codex · agents
+#   <platform> = opencode · cursor · windsurf · antigravity · pi · kiro · copilot · droid · dsh · gemini · codex · agents
 ```

 설치기는 스킬을 당신의 플랫폼에 맞는 폴더로 **평평하게** 복사하므로, 공유 프레임워크（`../_shared/`）가
@@ -297,12 +297,13 @@ curl -fsSL https://raw.githubusercontent.com/hyhmrright/brooks-lint/main/scripts
 | GitHub Copilot | `.github/skills`（`--project`） | `.claude/skills`, `AGENTS.md` | [설정](docs/copilot-setup.md) |
 | Kiro (AWS) | `~/.kiro/skills` | `AGENTS.md` | [설정](docs/kiro-setup.md) |
 | Factory Droid | `~/.factory/skills` | `AGENTS.md` | [설정](docs/factory-droid-setup.md) |
+| DeepSeek Harness (`dsh`) | `~/.dsh/skills` | `~/.agents/skills`, `AGENTS.md` | [설정](docs/dsh-setup.md) |

-Kiro와 Factory Droid는 `/brooks-review`도 자동 등록합니다. 스킬이 처음이거나 위 목록에 없는 에이전트를
-쓰시나요? **[docs/getting-started.md](docs/getting-started.md)**를 참고하세요.
+Kiro, Factory Droid, DeepSeek Harness는 `/brooks-review`도 자동 등록합니다. 스킬이 처음이거나 위
+목록에 없는 에이전트를 쓰시나요? **[docs/getting-started.md](docs/getting-started.md)**를 참고하세요.

 > **🧪 검증 상태.** Claude Code, Gemini CLI, Codex CLI는 메인테이너가 검증했습니다. 위
-> 여덟 개 플랫폼은 각 도구의 공식 스킬 명세를 토대로 문서화되었고 파일 레이아웃
+> 아홉 개 플랫폼은 각 도구의 공식 스킬 명세를 토대로 문서화되었고 파일 레이아웃
 > 수준에서 검증되었으나（설치기는 테스트되었음）, 메인테이너가 모든 플랫폼에서 end-to-end로 직접 실행해 보지는
 > 못했습니다. 어떤 것을 시도해 보셨나요 — 잘 되든 **안 되든**? 플랫폼, 버전, 본 결과를 담아
 > [이슈를 열어 주세요](https://github.com/hyhmrright/brooks-lint/issues/new).
@@ -322,9 +323,10 @@ Kiro와 Factory Droid는 `/brooks-review`도 자동 등록합니다. 스킬이

 **플랫폼별 문법.** Claude Code는 네임스페이스가 붙은 전체 형식 `/brooks-lint:brooks-review`도 받습니다
 — 짧은 형식은 session-start 훅이 첫 세션 시작 시 자동 설치합니다. Codex CLI는 `$brooks-review`를 씁니다.
-Gemini CLI는 위 표 그대로입니다. OpenCode, Cursor, Antigravity, pi는 각 스킬의 `description`에서 Agent
-Skills를 호출하므로 그냥 요청하면 됩니다（"이 PR을 리뷰해줘", "우리 최악의 기술 부채는 어디야?"）. 명시적
-호출이 필요하면 각 플랫폼의 문법을 쓰세요（pi는 각 스킬을 `/skill:brooks-review`로 등록）. 모든
+Gemini CLI는 위 표 그대로입니다. OpenCode, Cursor, Antigravity, pi, DeepSeek Harness는 각 스킬의
+`description`에서 Agent Skills를 호출하므로 그냥 요청하면 됩니다（"이 PR을 리뷰해줘", "우리 최악의 기술
+부채는 어디야?"）. 명시적 호출이 필요하면 각 플랫폼의 문법을 쓰세요（pi는 각 스킬을
+`/skill:brooks-review`로 등록; dsh는 위 표 그대로이며 `/` 메뉴에서 고르거나 직접 입력）. 모든
 플랫폼에서 코드 품질, 아키텍처, 테스트 건강을 이야기하면 스킬이 자동으로 트리거됩니다.

 > PR 리뷰에는 가벼운 Step 7 빠른 테스트 점검이 자동으로 포함됩니다（문서만 바뀐 diff에서는 건너뜀）.
@@ -446,6 +448,18 @@ jobs:

 `fail-on-regression`은 `.brooks-lint-history.json`을 읽으므로, "새로운 회귀 없음"을 강제하려면 그 파일을 커밋하세요. `sarif-file`을 설정하면 진단이 PR의 **Files changed** 탭에 인라인으로 나타나며, job에 `security-events: write` 권한이 필요합니다.

+**사용자 지정 API 엔드포인트.** `api-base-url`을 지정하면 action이 `api.anthropic.com` 대신 Anthropic 호환 `/v1/messages` 엔드포인트(자체 호스팅 프록시, LLM 게이트웨이, 리전 미러)를 호출합니다. 해당 엔드포인트의 키를 `anthropic-api-key`로, 기대하는 모델 id를 `model`로 전달하세요:
+
+```yaml
+        with:
+          mode: review
+          api-base-url: https://your-gateway.example.com
+          anthropic-api-key: ${{ secrets.GATEWAY_API_KEY }}
+          model: gateway-model-id
+```
+
+brooks-lint는 여기에 지정한 호스트로 diff를 전송하므로, 소스 코드를 맡겨도 되는 곳만 지정하세요. `scripts/ci-review.mjs`를 직접 실행할 때는 플래그가 필요 없습니다 — Anthropic SDK가 `ANTHROPIC_BASE_URL`을 직접 읽습니다.
+
 **비용:** PR 실행당 약 $0.05–0.15로, diff 크기와 모델에 따라 다릅니다. `pull_request` 이벤트에서만 실행할 것을 권장합니다.

 ## 로드맵
@@ -484,7 +498,7 @@ MIT License — 자세한 내용은 [LICENSE](LICENSE)를 참고하세요.

 ## Star 히스토리

-[![Star History Chart](https://api.star-history.com/svg?repos=hyhmrright/brooks-lint&type=Date)](https://star-history.com/#hyhmrright/brooks-lint&Date)
+[![Star History](assets/star-history.svg)](https://github.com/hyhmrright/brooks-lint/stargazers)

 ---

@@ -27,7 +27,7 @@
 </p>

 <p align="center">
-  <img src="https://img.shields.io/badge/version-1.4.3-blue.svg" alt="Version">
+  <img src="https://img.shields.io/badge/version-1.5.0-blue.svg" alt="Version">
  <img src="https://img.shields.io/badge/license-MIT-green.svg" alt="MIT License">
  <img src="https://img.shields.io/badge/Claude_Code-Plugin-blueviolet.svg" alt="Claude Code Plugin">
  <img src="https://img.shields.io/badge/Codex_CLI-Skill-orange.svg" alt="Codex CLI Skill">
@@ -78,7 +78,7 @@ Then just ask ("review this PR", "audit the architecture"), or run one of the si
 ([what each one does](#slash-commands)).

 Every finding comes back as **Symptom → Source → Consequence → Remedy** with a book citation and a
-0–100 Health Score. Full install options (8 more platforms) and CI/CD setup are [below](#installation).
+0–100 Health Score. Full install options (9 more platforms) and CI/CD setup are [below](#installation).

 ## The Twelve Books

@@ -272,7 +272,7 @@ Install the brooks-lint skill from hyhmrright/brooks-lint       # ask inside a C

 Or use the installer below: `./scripts/install.sh gemini` / `./scripts/install.sh codex`.

-### Every other platform — OpenCode · Cursor · Windsurf · Antigravity · pi · Copilot · Kiro · Factory Droid
+### Every other platform — OpenCode · Cursor · Windsurf · Antigravity · pi · Copilot · Kiro · Factory Droid · DeepSeek Harness

 brooks-lint ships as standard [Agent Skills](https://agentskills.io). **Any agent that loads Agent
 Skills runs all six modes with no conversion** — one command installs them:
@@ -280,7 +280,7 @@ Skills runs all six modes with no conversion** — one command installs them:
 ```bash
 # pick your platform; --project installs into the current repo instead of your global config
 curl -fsSL https://raw.githubusercontent.com/hyhmrright/brooks-lint/main/scripts/install.sh | bash -s -- <platform>
-#   <platform> = opencode · cursor · windsurf · antigravity · pi · kiro · copilot · droid · gemini · codex · agents
+#   <platform> = opencode · cursor · windsurf · antigravity · pi · kiro · copilot · droid · dsh · gemini · codex · agents
 ```

 The installer copies the skills **flat** into the right folder, so the shared framework
@@ -297,12 +297,13 @@ The installer copies the skills **flat** into the right folder, so the shared fr
 | GitHub Copilot | `.github/skills` (`--project`) | `.claude/skills`, `AGENTS.md` | [setup](docs/copilot-setup.md) |
 | Kiro (AWS) | `~/.kiro/skills` | `AGENTS.md` | [setup](docs/kiro-setup.md) |
 | Factory Droid | `~/.factory/skills` | `AGENTS.md` | [setup](docs/factory-droid-setup.md) |
+| DeepSeek Harness (`dsh`) | `~/.dsh/skills` | `~/.agents/skills`, `AGENTS.md` | [setup](docs/dsh-setup.md) |

-Kiro and Factory Droid also auto-register `/brooks-review`. New to skills, or using an agent not
-listed? See **[docs/getting-started.md](docs/getting-started.md)**.
+Kiro, Factory Droid, and DeepSeek Harness also auto-register `/brooks-review`. New to skills, or
+using an agent not listed? See **[docs/getting-started.md](docs/getting-started.md)**.

 > **🧪 Verification status.** Claude Code, Gemini CLI, and Codex CLI are maintainer-verified. The
-> eight platforms above are documented from each tool's official skill spec and verified at the
+> nine platforms above are documented from each tool's official skill spec and verified at the
 > file-layout level (the installer is tested), but not yet run end-to-end by the maintainer on every
 > platform. Tried one — working **or** broken?
 > [Open an issue](https://github.com/hyhmrright/brooks-lint/issues/new) with the platform, version,
@@ -323,9 +324,10 @@ listed? See **[docs/getting-started.md](docs/getting-started.md)**.
 **Syntax by platform.** Claude Code also accepts the namespaced form
 `/brooks-lint:brooks-review` — short forms are auto-installed on first session start by the
 session-start hook. Codex CLI uses `$brooks-review`. Gemini CLI uses the table as written.
-OpenCode, Cursor, Antigravity, and pi invoke Agent Skills from each skill's `description`, so
-just ask ("review this PR", "where's our worst tech debt?"); for explicit invocation use the
-platform's own syntax (pi registers each skill as `/skill:brooks-review`). On every platform the
+OpenCode, Cursor, Antigravity, pi, and DeepSeek Harness invoke Agent Skills from each skill's
+`description`, so just ask ("review this PR", "where's our worst tech debt?"); for explicit
+invocation use the platform's own syntax (pi registers each skill as `/skill:brooks-review`; dsh
+takes the table as written, from its `/` menu or typed inline). On every platform the
 skills also trigger automatically when you discuss code quality, architecture, or test health.

 > PR reviews include a lightweight Step 7 Quick Test Check automatically (skipped for docs-only
@@ -448,6 +450,18 @@ The action posts the review as a PR comment and optionally fails the check if th

 `fail-on-regression` reads `.brooks-lint-history.json`, so commit that file to enforce "no new regressions". Setting `sarif-file` makes findings appear inline on the PR's **Files changed** tab and requires `security-events: write` permission on the job.

+**Custom API endpoint.** `api-base-url` points the action at any Anthropic-compatible `/v1/messages` endpoint — a self-hosted proxy, an LLM gateway, a regional mirror — instead of `api.anthropic.com`. Pass that endpoint's key as `anthropic-api-key` and the model id it expects as `model`:
+
+```yaml
+        with:
+          mode: review
+          api-base-url: https://your-gateway.example.com
+          anthropic-api-key: ${{ secrets.GATEWAY_API_KEY }}
+          model: gateway-model-id
+```
+
+brooks-lint sends your diff to whatever host you name here, so only point it at one you trust with your source. Running `scripts/ci-review.mjs` yourself needs no flag at all — the Anthropic SDK reads `ANTHROPIC_BASE_URL` directly.
+
 **Cost:** ~$0.05–0.15 per PR run depending on diff size and model. Recommend running on `pull_request` events only.

 ## Roadmap
@@ -486,7 +500,7 @@ their ideas, applied to modern code quality assessment.

 ## Star History

-[![Star History Chart](https://api.star-history.com/svg?repos=hyhmrright/brooks-lint&type=Date)](https://star-history.com/#hyhmrright/brooks-lint&Date)
+[![Star History](assets/star-history.svg)](https://github.com/hyhmrright/brooks-lint/stargazers)

 ---

@@ -27,7 +27,7 @@
 </p>

 <p align="center">
-  <img src="https://img.shields.io/badge/version-1.4.3-blue.svg" alt="Version">
+  <img src="https://img.shields.io/badge/version-1.5.0-blue.svg" alt="Version">
  <img src="https://img.shields.io/badge/license-MIT-green.svg" alt="MIT License">
  <img src="https://img.shields.io/badge/Claude_Code-Plugin-blueviolet.svg" alt="Claude Code Plugin">
  <img src="https://img.shields.io/badge/Codex_CLI-Skill-orange.svg" alt="Codex CLI Skill">
@@ -76,7 +76,7 @@ curl -fsSL https://raw.githubusercontent.com/hyhmrright/brooks-lint/main/scripts
 装好后直接开口（"审查这个 PR""审计架构"），或运行六个命令之一——`/brooks-review`、`/brooks-audit`、
 `/brooks-debt`、`/brooks-test`、`/brooks-health`、`/brooks-sweep`（[各自的作用](#斜杠命令)）。

-每条诊断都以 **症状 → 根源 → 后果 → 对策** 返回，附书目出处和 0–100 健康分。完整安装方式（另外 8 个
+每条诊断都以 **症状 → 根源 → 后果 → 对策** 返回，附书目出处和 0–100 健康分。完整安装方式（另外 9 个
 平台）和 CI/CD 配置见[下文](#安装)。

 ## 十二本书
@@ -270,7 +270,7 @@ Install the brooks-lint skill from hyhmrright/brooks-lint       # 在 Codex 会

 或使用下面的安装器：`./scripts/install.sh gemini` / `./scripts/install.sh codex`。

-### 其它所有平台——OpenCode · Cursor · Windsurf · Antigravity · pi · Copilot · Kiro · Factory Droid
+### 其它所有平台——OpenCode · Cursor · Windsurf · Antigravity · pi · Copilot · Kiro · Factory Droid · DeepSeek Harness

 brooks-lint 以标准 [Agent Skills](https://agentskills.io) 形式分发。**任何加载 Agent Skills 的 agent
 都能无需任何转换运行全部六种模式**——一条命令即可安装：
@@ -278,7 +278,7 @@ brooks-lint 以标准 [Agent Skills](https://agentskills.io) 形式分发。**
 ```bash
 # 选择你的平台；加 --project 装进当前仓库而非全局配置
 curl -fsSL https://raw.githubusercontent.com/hyhmrright/brooks-lint/main/scripts/install.sh | bash -s -- <平台>
-#   <平台> = opencode · cursor · windsurf · antigravity · pi · kiro · copilot · droid · gemini · codex · agents
+#   <平台> = opencode · cursor · windsurf · antigravity · pi · kiro · copilot · droid · dsh · gemini · codex · agents
 ```

 安装器会把技能**扁平**拷进该平台对应的文件夹，让共享框架（`../_shared/`）始终正确解析——你不可能装错布局。
@@ -294,11 +294,12 @@ curl -fsSL https://raw.githubusercontent.com/hyhmrright/brooks-lint/main/scripts
 | GitHub Copilot | `.github/skills`（`--project`） | `.claude/skills`、`AGENTS.md` | [配置](docs/copilot-setup.md) |
 | Kiro（AWS） | `~/.kiro/skills` | `AGENTS.md` | [配置](docs/kiro-setup.md) |
 | Factory Droid | `~/.factory/skills` | `AGENTS.md` | [配置](docs/factory-droid-setup.md) |
+| DeepSeek Harness（`dsh`） | `~/.dsh/skills` | `~/.agents/skills`、`AGENTS.md` | [配置](docs/dsh-setup.md) |

-Kiro 与 Factory Droid 还会自动注册 `/brooks-review`。不熟悉 skills、或用的是上面没列出的 agent？
-见 **[docs/getting-started.md](docs/getting-started.md)**。
+Kiro、Factory Droid 与 DeepSeek Harness 还会自动注册 `/brooks-review`。不熟悉 skills、或用的是上面
+没列出的 agent？见 **[docs/getting-started.md](docs/getting-started.md)**。

-> **🧪 验证状态。** Claude Code、Gemini CLI、Codex CLI 已由维护者验证。上面八个平台依据各工具官方技能规范编写，
+> **🧪 验证状态。** Claude Code、Gemini CLI、Codex CLI 已由维护者验证。上面九个平台依据各工具官方技能规范编写，
 > 并已在文件布局层面验证（安装器经过测试），但维护者尚未在每个平台端到端实跑。在某平台试过了——无论成功**还是**失败？
 > 请[提一个 issue](https://github.com/hyhmrright/brooks-lint/issues/new)，附上平台、版本和你看到的结果。
 > 用的是其它兼容 Agent Skills 的 agent？它几乎肯定以同样方式工作——告诉我们，我们会补上。
@@ -316,9 +317,10 @@ Kiro 与 Factory Droid 还会自动注册 `/brooks-review`。不熟悉 skills、

 **各平台语法。** Claude Code 也接受带命名空间的完整形式 `/brooks-lint:brooks-review`——短命令由
 session-start 钩子在首次会话启动时自动安装。Codex CLI 用 `$brooks-review`。Gemini CLI 直接用上表。
-OpenCode、Cursor、Antigravity、pi 依据每个技能的 `description` 自动调用 Agent Skills，直接提问即可
-（"审查这个 PR"、"我们最糟的技术债在哪"）；需要显式调用时用各平台自己的语法（pi 把每个技能注册为
-`/skill:brooks-review`）。在所有平台上，当你讨论代码质量、架构或测试健康时，这些技能也会自动触发。
+OpenCode、Cursor、Antigravity、pi、DeepSeek Harness 依据每个技能的 `description` 自动调用 Agent
+Skills，直接提问即可（"审查这个 PR"、"我们最糟的技术债在哪"）；需要显式调用时用各平台自己的语法
+（pi 把每个技能注册为 `/skill:brooks-review`；dsh 直接用上表，可从 `/` 菜单选或手打）。在所有平台上，
+当你讨论代码质量、架构或测试健康时，这些技能也会自动触发。

 > PR 审查会自动包含一个轻量的第 7 步快速测试检查（对纯文档 diff 会跳过）。需要完整的测试审查请用
 > `/brooks-test`；需要某个维度的深度诊断时，请用该维度的专项技能，而不是 `/brooks-health`。
@@ -437,6 +439,18 @@ jobs:

 `fail-on-regression` 读取 `.brooks-lint-history.json`，因此提交该文件即可强制"无新增回归"。设置 `sarif-file` 会让诊断直接显示在 PR 的 **Files changed** 标签页，并需要 job 具备 `security-events: write` 权限。

+**自定义 API 端点。** `api-base-url` 让 action 改为调用任意 Anthropic 兼容的 `/v1/messages` 端点 —— 自建代理、LLM 网关或区域镜像 —— 而不是 `api.anthropic.com`。把该端点的密钥作为 `anthropic-api-key` 传入，把它期望的模型 id 作为 `model` 传入：
+
+```yaml
+        with:
+          mode: review
+          api-base-url: https://your-gateway.example.com
+          anthropic-api-key: ${{ secrets.GATEWAY_API_KEY }}
+          model: gateway-model-id
+```
+
+brooks-lint 会把你的 diff 发送到这里填写的主机，因此只应指向你信任其接触源码的一方。自己运行 `scripts/ci-review.mjs` 完全不需要参数 —— Anthropic SDK 会直接读取 `ANTHROPIC_BASE_URL`。
+
 **成本：** 每次 PR 运行约 $0.05–0.15，取决于 diff 大小和模型。建议仅在 `pull_request` 事件上运行。

 ## 路线图
@@ -472,7 +486,7 @@ MIT License——详见 [LICENSE](LICENSE)。

 ## Star 历史

-[![Star History Chart](https://api.star-history.com/svg?repos=hyhmrright/brooks-lint&type=Date)](https://star-history.com/#hyhmrright/brooks-lint&Date)
+[![Star History](assets/star-history.svg)](https://github.com/hyhmrright/brooks-lint/stargazers)

 ---

@@ -27,7 +27,7 @@
 </p>

 <p align="center">
-  <img src="https://img.shields.io/badge/version-1.4.3-blue.svg" alt="Version">
+  <img src="https://img.shields.io/badge/version-1.5.0-blue.svg" alt="Version">
  <img src="https://img.shields.io/badge/license-MIT-green.svg" alt="MIT License">
  <img src="https://img.shields.io/badge/Claude_Code-Plugin-blueviolet.svg" alt="Claude Code Plugin">
  <img src="https://img.shields.io/badge/Codex_CLI-Skill-orange.svg" alt="Codex CLI Skill">
@@ -76,7 +76,7 @@ curl -fsSL https://raw.githubusercontent.com/hyhmrright/brooks-lint/main/scripts
 裝好後直接開口（「審查這個 PR」「稽核架構」），或執行六個命令之一——`/brooks-review`、`/brooks-audit`、
 `/brooks-debt`、`/brooks-test`、`/brooks-health`、`/brooks-sweep`（[各自的作用](#斜線命令)）。

-每條診斷都以 **症狀 → 根源 → 後果 → 對策** 回傳，附書目出處和 0–100 健康分。完整安裝方式（另外 8 個
+每條診斷都以 **症狀 → 根源 → 後果 → 對策** 回傳，附書目出處和 0–100 健康分。完整安裝方式（另外 9 個
 平台）和 CI/CD 設定見[下文](#安裝)。

 ## 十二本書
@@ -270,7 +270,7 @@ Install the brooks-lint skill from hyhmrright/brooks-lint       # 在 Codex 工

 或使用下面的安裝器：`./scripts/install.sh gemini` / `./scripts/install.sh codex`。

-### 其他所有平台——OpenCode · Cursor · Windsurf · Antigravity · pi · Copilot · Kiro · Factory Droid
+### 其他所有平台——OpenCode · Cursor · Windsurf · Antigravity · pi · Copilot · Kiro · Factory Droid · DeepSeek Harness

 brooks-lint 以標準 [Agent Skills](https://agentskills.io) 形式散布。**任何載入 Agent Skills 的 agent
 都能無需任何轉換執行全部六種模式**——一條命令即可安裝：
@@ -278,7 +278,7 @@ brooks-lint 以標準 [Agent Skills](https://agentskills.io) 形式散布。**
 ```bash
 # 選擇你的平台；加 --project 裝進當前儲存庫而非全域設定
 curl -fsSL https://raw.githubusercontent.com/hyhmrright/brooks-lint/main/scripts/install.sh | bash -s -- <平台>
-#   <平台> = opencode · cursor · windsurf · antigravity · pi · kiro · copilot · droid · gemini · codex · agents
+#   <平台> = opencode · cursor · windsurf · antigravity · pi · kiro · copilot · droid · dsh · gemini · codex · agents
 ```

 安裝器會把技能**扁平**複製進該平台對應的資料夾，讓共享框架（`../_shared/`）始終正確解析——你不可能裝錯佈局。
@@ -294,11 +294,12 @@ curl -fsSL https://raw.githubusercontent.com/hyhmrright/brooks-lint/main/scripts
 | GitHub Copilot | `.github/skills`（`--project`） | `.claude/skills`、`AGENTS.md` | [設定](docs/copilot-setup.md) |
 | Kiro（AWS） | `~/.kiro/skills` | `AGENTS.md` | [設定](docs/kiro-setup.md) |
 | Factory Droid | `~/.factory/skills` | `AGENTS.md` | [設定](docs/factory-droid-setup.md) |
+| DeepSeek Harness（`dsh`） | `~/.dsh/skills` | `~/.agents/skills`、`AGENTS.md` | [設定](docs/dsh-setup.md) |

-Kiro 與 Factory Droid 還會自動註冊 `/brooks-review`。不熟悉 skills、或用的是上面沒列出的 agent？
-見 **[docs/getting-started.md](docs/getting-started.md)**。
+Kiro、Factory Droid 與 DeepSeek Harness 還會自動註冊 `/brooks-review`。不熟悉 skills、或用的是上面
+沒列出的 agent？見 **[docs/getting-started.md](docs/getting-started.md)**。

-> **🧪 驗證狀態。** Claude Code、Gemini CLI、Codex CLI 已由維護者驗證。上面八個平台依據各工具官方技能規範撰寫，
+> **🧪 驗證狀態。** Claude Code、Gemini CLI、Codex CLI 已由維護者驗證。上面九個平台依據各工具官方技能規範撰寫，
 > 並已在檔案佈局層面驗證（安裝器經過測試），但維護者尚未在每個平台端到端實跑。在某平台試過了——無論成功**還是**失敗？
 > 請[提一個 issue](https://github.com/hyhmrright/brooks-lint/issues/new)，附上平台、版本和你看到的結果。
 > 用的是其他相容 Agent Skills 的 agent？它幾乎肯定以同樣方式運作——告訴我們，我們會補上。
@@ -316,9 +317,10 @@ Kiro 與 Factory Droid 還會自動註冊 `/brooks-review`。不熟悉 skills、

 **各平台語法。** Claude Code 也接受帶命名空間的完整形式 `/brooks-lint:brooks-review`——短命令由
 session-start 鉤子在首次工作階段啟動時自動安裝。Codex CLI 用 `$brooks-review`。Gemini CLI 直接用上表。
-OpenCode、Cursor、Antigravity、pi 依據每個技能的 `description` 自動呼叫 Agent Skills，直接提問即可
-（「審查這個 PR」、「我們最糟的技術債在哪」）；需要顯式呼叫時用各平台自己的語法（pi 把每個技能註冊為
-`/skill:brooks-review`）。在所有平台上，當你討論程式碼品質、架構或測試健康時，這些技能也會自動觸發。
+OpenCode、Cursor、Antigravity、pi、DeepSeek Harness 依據每個技能的 `description` 自動呼叫 Agent
+Skills，直接提問即可（「審查這個 PR」、「我們最糟的技術債在哪」）；需要顯式呼叫時用各平台自己的語法
+（pi 把每個技能註冊為 `/skill:brooks-review`；dsh 直接用上表，可從 `/` 選單挑或手打）。在所有平台上，
+當你討論程式碼品質、架構或測試健康時，這些技能也會自動觸發。

 > PR 審查會自動包含一個輕量的第 7 步快速測試檢查（對純文件 diff 會跳過）。需要完整的測試稽核請用
 > `/brooks-test`；需要某個維度的深度診斷時，請用該維度的專項技能，而不是 `/brooks-health`。
@@ -437,6 +439,18 @@ jobs:

 `fail-on-regression` 讀取 `.brooks-lint-history.json`，因此提交該檔案即可強制「無新增回歸」。設定 `sarif-file` 會讓診斷直接顯示在 PR 的 **Files changed** 分頁，並需要 job 具備 `security-events: write` 權限。

+**自訂 API 端點。** `api-base-url` 讓 action 改為呼叫任意 Anthropic 相容的 `/v1/messages` 端點 —— 自建代理、LLM 閘道或區域鏡像 —— 而不是 `api.anthropic.com`。把該端點的金鑰作為 `anthropic-api-key` 傳入，把它預期的模型 id 作為 `model` 傳入：
+
+```yaml
+        with:
+          mode: review
+          api-base-url: https://your-gateway.example.com
+          anthropic-api-key: ${{ secrets.GATEWAY_API_KEY }}
+          model: gateway-model-id
+```
+
+brooks-lint 會把你的 diff 傳送到這裡填寫的主機，因此只應指向你信任其接觸原始碼的一方。自己執行 `scripts/ci-review.mjs` 完全不需要參數 —— Anthropic SDK 會直接讀取 `ANTHROPIC_BASE_URL`。
+
 **成本：** 每次 PR 執行約 $0.05–0.15，取決於 diff 大小和模型。建議僅在 `pull_request` 事件上執行。

 ## 路線圖
@@ -472,7 +486,7 @@ MIT License——詳見 [LICENSE](LICENSE)。

 ## Star 歷史

-[![Star History Chart](https://api.star-history.com/svg?repos=hyhmrright/brooks-lint&type=Date)](https://star-history.com/#hyhmrright/brooks-lint&Date)
+[![Star History](assets/star-history.svg)](https://github.com/hyhmrright/brooks-lint/stargazers)

 ---

@@ -1,8 +1,8 @@
 # Source

 - Repo: https://github.com/hyhmrright/brooks-lint
- Ref: 814174cd5b340bc0d8b0161b6d8288980428a44d
+- Ref: e9acf5c6100d2d786ed58f0fb7b9566a6218926a
 - Remove-Paths: 
- Snapshot: 2026-08-12
+- Snapshot: 2026-08-24
 - Sync-Mode: copy_skill_dirs
 - Notes: vendored into playbook branch thirdparty/skill
@@ -0,0 +1,45 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="800" height="400" viewBox="0 0 800 400" role="img" aria-label="Star history for hyhmrright/brooks-lint: 1393 stars">
+  <style>
+    .bg { fill: #ffffff; }
+    .title { fill: #111827; font: 600 16px -apple-system, BlinkMacSystemFont, "Segoe UI", Helvetica, Arial, sans-serif; }
+    .sub { fill: #6b7280; font: 400 12px -apple-system, BlinkMacSystemFont, "Segoe UI", Helvetica, Arial, sans-serif; }
+    .tick { fill: #6b7280; font: 400 11px -apple-system, BlinkMacSystemFont, "Segoe UI", Helvetica, Arial, sans-serif; }
+    .grid { stroke: #e5e7eb; stroke-width: 1; }
+    .axis { stroke: #d1d5db; stroke-width: 1; }
+    .total { fill: #3b82f6; font: 600 13px -apple-system, BlinkMacSystemFont, "Segoe UI", Helvetica, Arial, sans-serif; }
+    @media (prefers-color-scheme: dark) {
+      .bg { fill: #0d1117; }
+      .title { fill: #e6edf3; }
+      .sub, .tick { fill: #8b949e; }
+      .grid { stroke: #21262d; }
+      .axis { stroke: #30363d; }
+    }
+  </style>
+  <defs>
+    <linearGradient id="fade" x1="0" y1="0" x2="0" y2="1">
+      <stop offset="0%" stop-color="#3b82f6" stop-opacity="0.28"/>
+      <stop offset="100%" stop-color="#3b82f6" stop-opacity="0.02"/>
+    </linearGradient>
+  </defs>
+  <rect width="800" height="400" class="bg"/>
+  <text x="64" y="26" class="title">Star History</text>
+  <text x="776" y="26" class="sub" text-anchor="end">hyhmrright/brooks-lint</text>
+  <g>
+    <line x1="64" y1="352.0" x2="776" y2="352.0" class="grid"/><text x="54" y="356.0" class="tick" text-anchor="end">0</text>
+    <line x1="64" y1="249.3" x2="776" y2="249.3" class="grid"/><text x="54" y="253.3" class="tick" text-anchor="end">500</text>
+    <line x1="64" y1="146.7" x2="776" y2="146.7" class="grid"/><text x="54" y="150.7" class="tick" text-anchor="end">1,000</text>
+    <line x1="64" y1="44.0" x2="776" y2="44.0" class="grid"/><text x="54" y="48.0" class="tick" text-anchor="end">1,500</text>
+  </g>
+  <line x1="64" y1="352" x2="776" y2="352" class="axis"/>
+  <path d="M64.0 351.8L82.8 350.8L92.5 349.7L99.2 348.7L116.3 347.7L125.3 346.7L156.4 345.6L168.8 344.6L178.1 343.6L186.6 342.6L194.8 341.5L198.1 340.5L205.9 339.5L212.1 338.4L215.9 337.4L218.4 336.4L222.5 335.4L227.0 334.3L229.0 333.3L239.8 332.3L246.3 331.3L257.3 330.2L259.1 329.2L268.1 328.2L279.8 327.2L290.7 326.1L296.3 325.1L303.7 324.1L312.2 323.0L318.6 322.0L322.8 321.0L330.8 320.0L344.8 318.9L349.9 317.9L355.5 316.9L360.2 315.9L372.6 314.8L386.8 313.8L404.4 312.8L413.5 311.8L413.6 310.7L413.7 309.7L413.8 308.7L413.8 307.6L413.8 306.6L413.8 305.6L413.9 304.6L413.9 303.5L413.9 302.5L413.9 301.5L414.0 300.5L414.0 299.4L414.0 298.4L414.1 297.4L414.1 296.4L414.1 295.3L414.2 294.3L414.2 293.3L414.2 292.2L414.3 291.2L414.3 290.2L414.4 289.2L414.4 288.1L414.5 287.1L414.5 286.1L414.5 285.1L414.6 284.0L414.6 283.0L414.6 282.0L414.7 281.0L414.7 279.9L414.7 278.9L414.8 277.9L414.8 276.8L414.8 275.8L414.8 274.8L414.9 273.8L414.9 272.7L414.9 271.7L414.9 270.7L414.9 269.7L414.9 268.6L415.0 267.6L415.0 266.6L415.0 265.6L415.0 264.5L415.1 263.5L415.1 262.5L415.1 261.4L415.2 260.4L415.2 259.4L415.3 258.4L415.3 257.3L415.3 256.3L415.3 255.3L415.4 254.3L415.4 253.2L415.4 252.2L415.4 251.2L415.5 250.2L415.5 249.1L415.5 248.1L415.6 247.1L415.6 246.0L415.6 245.0L415.6 244.0L415.7 243.0L415.7 241.9L415.7 240.9L415.7 239.9L415.7 238.9L415.8 237.8L415.8 236.8L415.8 235.8L415.8 234.8L415.8 233.7L415.9 232.7L415.9 231.7L415.9 230.6L416.0 229.6L416.0 228.6L416.1 227.6L416.2 226.5L416.2 225.5L416.3 224.5L416.4 223.5L416.6 222.4L416.8 221.4L416.9 220.4L417.0 219.4L417.1 218.3L417.1 217.3L417.2 216.3L417.3 215.2L417.4 214.2L417.4 213.2L417.5 212.2L417.5 211.1L417.6 210.1L417.7 209.1L417.7 208.1L417.8 207.0L417.9 206.0L417.9 205.0L418.0 204.0L418.1 202.9L418.2 201.9L418.3 200.9L418.4 199.8L418.6 198.8L418.7 197.8L418.8 196.8L418.9 195.7L419.0 194.7L419.1 193.7L419.2 192.7L419.4 191.6L419.5 190.6L419.5 189.6L419.6 188.6L419.7 187.5L419.8 186.5L420.0 185.5L420.1 184.4L420.2 183.4L420.3 182.4L420.4 181.4L420.5 180.3L420.6 179.3L420.8 178.3L421.2 177.3L421.9 176.2L422.6 175.2L423.4 174.2L423.6 173.2L424.1 172.1L425.9 171.1L427.4 170.1L428.8 169.0L430.5 168.0L434.1 167.0L437.1 166.0L438.3 164.9L439.4 163.9L442.1 162.9L442.2 161.9L442.3 160.8L442.4 159.8L442.6 158.8L442.7 157.8L442.8 156.7L443.0 155.7L443.2 154.7L443.6 153.6L443.7 152.6L443.8 151.6L444.1 150.6L444.3 149.5L444.5 148.5L444.6 147.5L444.7 146.5L444.8 145.4L444.9 144.4L444.9 143.4L445.3 142.4L445.9 141.3L446.5 140.3L446.7 139.3L447.2 138.2L447.7 137.2L448.4 136.2L452.1 135.2L452.7 134.1L453.3 133.1L454.4 132.1L457.6 131.1L460.0 130.0L462.9 129.0L463.8 128.0L468.1 127.0L473.2 125.9L478.9 124.9L482.0 123.9L485.9 122.8L492.4 121.8L493.5 120.8L501.3 119.8L505.4 118.7L508.4 117.7L517.8 116.7L520.3 115.7L523.7 114.6L527.5 113.6L531.8 112.6L535.5 111.6L542.7 110.5L551.8 109.5L559.4 108.5L561.3 107.4L565.7 106.4L575.1 105.4L578.5 104.4L587.9 103.3L589.5 102.3L593.6 101.3L604.0 100.3L620.8 99.2L631.3 98.2L634.9 97.2L656.5 96.2L658.1 95.1L662.9 94.1L671.4 93.1L672.3 92.0L675.8 91.0L687.6 90.0L698.1 89.0L717.6 87.9L725.3 86.9L732.2 85.9L747.1 84.9L747.6 83.8L748.3 82.8L749.0 81.8L749.5 80.8L750.2 79.7L750.8 78.7L751.1 77.7L751.8 76.6L752.6 75.6L754.4 74.6L755.3 73.6L756.8 72.5L760.1 71.5L762.4 70.5L765.0 69.5L769.5 68.4L770.7 67.4L774.8 66.4L776.0 66.0L776.0 352.0L64.0 352.0Z" fill="url(#fade)"/>
+  <path d="M64.0 351.8L82.8 350.8L92.5 349.7L99.2 348.7L116.3 347.7L125.3 346.7L156.4 345.6L168.8 344.6L178.1 343.6L186.6 342.6L194.8 341.5L198.1 340.5L205.9 339.5L212.1 338.4L215.9 337.4L218.4 336.4L222.5 335.4L227.0 334.3L229.0 333.3L239.8 332.3L246.3 331.3L257.3 330.2L259.1 329.2L268.1 328.2L279.8 327.2L290.7 326.1L296.3 325.1L303.7 324.1L312.2 323.0L318.6 322.0L322.8 321.0L330.8 320.0L344.8 318.9L349.9 317.9L355.5 316.9L360.2 315.9L372.6 314.8L386.8 313.8L404.4 312.8L413.5 311.8L413.6 310.7L413.7 309.7L413.8 308.7L413.8 307.6L413.8 306.6L413.8 305.6L413.9 304.6L413.9 303.5L413.9 302.5L413.9 301.5L414.0 300.5L414.0 299.4L414.0 298.4L414.1 297.4L414.1 296.4L414.1 295.3L414.2 294.3L414.2 293.3L414.2 292.2L414.3 291.2L414.3 290.2L414.4 289.2L414.4 288.1L414.5 287.1L414.5 286.1L414.5 285.1L414.6 284.0L414.6 283.0L414.6 282.0L414.7 281.0L414.7 279.9L414.7 278.9L414.8 277.9L414.8 276.8L414.8 275.8L414.8 274.8L414.9 273.8L414.9 272.7L414.9 271.7L414.9 270.7L414.9 269.7L414.9 268.6L415.0 267.6L415.0 266.6L415.0 265.6L415.0 264.5L415.1 263.5L415.1 262.5L415.1 261.4L415.2 260.4L415.2 259.4L415.3 258.4L415.3 257.3L415.3 256.3L415.3 255.3L415.4 254.3L415.4 253.2L415.4 252.2L415.4 251.2L415.5 250.2L415.5 249.1L415.5 248.1L415.6 247.1L415.6 246.0L415.6 245.0L415.6 244.0L415.7 243.0L415.7 241.9L415.7 240.9L415.7 239.9L415.7 238.9L415.8 237.8L415.8 236.8L415.8 235.8L415.8 234.8L415.8 233.7L415.9 232.7L415.9 231.7L415.9 230.6L416.0 229.6L416.0 228.6L416.1 227.6L416.2 226.5L416.2 225.5L416.3 224.5L416.4 223.5L416.6 222.4L416.8 221.4L416.9 220.4L417.0 219.4L417.1 218.3L417.1 217.3L417.2 216.3L417.3 215.2L417.4 214.2L417.4 213.2L417.5 212.2L417.5 211.1L417.6 210.1L417.7 209.1L417.7 208.1L417.8 207.0L417.9 206.0L417.9 205.0L418.0 204.0L418.1 202.9L418.2 201.9L418.3 200.9L418.4 199.8L418.6 198.8L418.7 197.8L418.8 196.8L418.9 195.7L419.0 194.7L419.1 193.7L419.2 192.7L419.4 191.6L419.5 190.6L419.5 189.6L419.6 188.6L419.7 187.5L419.8 186.5L420.0 185.5L420.1 184.4L420.2 183.4L420.3 182.4L420.4 181.4L420.5 180.3L420.6 179.3L420.8 178.3L421.2 177.3L421.9 176.2L422.6 175.2L423.4 174.2L423.6 173.2L424.1 172.1L425.9 171.1L427.4 170.1L428.8 169.0L430.5 168.0L434.1 167.0L437.1 166.0L438.3 164.9L439.4 163.9L442.1 162.9L442.2 161.9L442.3 160.8L442.4 159.8L442.6 158.8L442.7 157.8L442.8 156.7L443.0 155.7L443.2 154.7L443.6 153.6L443.7 152.6L443.8 151.6L444.1 150.6L444.3 149.5L444.5 148.5L444.6 147.5L444.7 146.5L444.8 145.4L444.9 144.4L444.9 143.4L445.3 142.4L445.9 141.3L446.5 140.3L446.7 139.3L447.2 138.2L447.7 137.2L448.4 136.2L452.1 135.2L452.7 134.1L453.3 133.1L454.4 132.1L457.6 131.1L460.0 130.0L462.9 129.0L463.8 128.0L468.1 127.0L473.2 125.9L478.9 124.9L482.0 123.9L485.9 122.8L492.4 121.8L493.5 120.8L501.3 119.8L505.4 118.7L508.4 117.7L517.8 116.7L520.3 115.7L523.7 114.6L527.5 113.6L531.8 112.6L535.5 111.6L542.7 110.5L551.8 109.5L559.4 108.5L561.3 107.4L565.7 106.4L575.1 105.4L578.5 104.4L587.9 103.3L589.5 102.3L593.6 101.3L604.0 100.3L620.8 99.2L631.3 98.2L634.9 97.2L656.5 96.2L658.1 95.1L662.9 94.1L671.4 93.1L672.3 92.0L675.8 91.0L687.6 90.0L698.1 89.0L717.6 87.9L725.3 86.9L732.2 85.9L747.1 84.9L747.6 83.8L748.3 82.8L749.0 81.8L749.5 80.8L750.2 79.7L750.8 78.7L751.1 77.7L751.8 76.6L752.6 75.6L754.4 74.6L755.3 73.6L756.8 72.5L760.1 71.5L762.4 70.5L765.0 69.5L769.5 68.4L770.7 67.4L774.8 66.4L776.0 66.0" fill="none" stroke="#3b82f6" stroke-width="2.5" stroke-linejoin="round" stroke-linecap="round"/>
+  <circle cx="776.0" cy="66.0" r="4" fill="#3b82f6"/>
+  <text x="764.0" y="70.0" class="total" text-anchor="end">1,393</text>
+  <g>
+    <text x="85.7" y="374" class="tick" text-anchor="middle">Apr 2026</text>
+    <text x="232.0" y="374" class="tick" text-anchor="middle">May</text>
+    <text x="383.1" y="374" class="tick" text-anchor="middle">Jun</text>
+    <text x="529.4" y="374" class="tick" text-anchor="middle">Jul</text>
+    <text x="680.6" y="374" class="tick" text-anchor="middle">Aug</text>
+  </g>
+</svg>
@@ -0,0 +1,60 @@
+# DeepSeek Harness (dsh) Setup
+
+[DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) (`dsh`) is DeepSeek AI's
+open-source agent harness — an "everything is a plugin" architecture with a Web UI, started with
+`npx @deepseek-ai/dsh web`. It natively loads [Agent Skills](https://agentskills.io) and reads
+`AGENTS.md`, so all six brooks-lint modes run with no conversion.
+
+## Install
+
+```bash
+# simplest — one command (global)
+curl -fsSL https://raw.githubusercontent.com/hyhmrright/brooks-lint/main/scripts/install.sh | bash -s -- dsh
+
+# from a clone
+./scripts/install.sh dsh            # global: ~/.dsh/skills (or $DSH_HOME/skills)
+./scripts/install.sh dsh --project  # this repo: ./.dsh/skills
+```
+
+Prefer a manual copy? Clone the repo and `cp -r skills/* ~/.dsh/skills/` — the contents, not the
+`skills/` folder itself, so `_shared/` lands as a sibling of the `brooks-*` folders.
+
+dsh scans these skill roots, highest priority first, so an existing vendor-neutral install is picked
+up automatically:
+
+| Root | Notes |
+|---|---|
+| `<projectRoot>/.dsh/skills` | what `--project` writes |
+| `<projectRoot>/.agents/skills` | shared with Cursor, Copilot, pi |
+| `$DSH_HOME/skills` (default `~/.dsh/skills`) | what the global install writes |
+| `$DSH_AGENTS_HOME/skills` (default `~/.agents/skills`) | `./scripts/install.sh agents` also covers dsh |
+
+The project root is the nearest ancestor containing `.git`; without one, dsh uses the current
+directory. When the same skill name appears in two roots, the higher one wins.
+
+## Invoke
+
+Just ask — dsh routes to a skill from its `description`:
+
+- "review this PR" → `brooks-review`
+- "audit the architecture" → `brooks-audit`
+- "where's our worst tech debt?" → `brooks-debt`
+
+For explicit invocation, type `/` in the Web UI prompt and pick the skill, or type the token by hand:
+`/brooks-review`, `/brooks-audit`, `/brooks-debt`, `/brooks-test`, `/brooks-health`, `/brooks-sweep`.
+dsh recognises a whitespace-bounded `/name` anywhere in a message and injects that skill's body
+deterministically. The repo's `AGENTS.md` carries the Iron Law (Symptom → Source → Consequence →
+Remedy) and the Health Score rules; dsh also loads `$DSH_HOME/AGENTS.md` plus every `AGENTS.md` from
+the project root down to your working directory.
+
+## Notes
+
+- **Flat layout** is mandatory (the installer guarantees it): discovery is one level deep
+  (`<root>/<name>/SKILL.md`), never recursive, and the skills read `../_shared/`, which only resolves
+  when `_shared/` sits beside the `brooks-*` folders. `_shared/` itself has no `SKILL.md`, so dsh
+  ignores it as a skill and the modes read it as ordinary files.
+- dsh is in developer preview and warns of compatibility-breaking changes; the skill discovery
+  contract above reflects `packages/skill/skill-filesystem` as of August 2026.
+- 🧪 Documented per the official [skills subsystem](https://github.com/deepseek-ai/deepseek-harness/blob/master/docs/subsystems/skills.md)
+  docs; community end-to-end verification welcome —
+  [open an issue](https://github.com/hyhmrright/brooks-lint/issues/new).
@@ -33,10 +33,10 @@ layout wrong:
 curl -fsSL https://raw.githubusercontent.com/hyhmrright/brooks-lint/main/scripts/install.sh | bash -s -- <platform>
 ```

-`<platform>` ∈ `opencode · cursor · windsurf · antigravity · pi · kiro · copilot · droid · gemini ·
-codex · claude · agents`. Add `--project` to install into the current repo instead of your global
-config. `agents` targets the vendor-neutral `~/.agents/skills` folder that Cursor, Copilot, pi, Gemini,
-and Codex all read.
+`<platform>` ∈ `opencode · cursor · windsurf · antigravity · pi · kiro · copilot · droid · dsh ·
+gemini · codex · claude · agents`. Add `--project` to install into the current repo instead of your
+global config. `agents` targets the vendor-neutral `~/.agents/skills` folder that Cursor, Copilot, pi,
+Gemini, Codex, and DeepSeek Harness all read.

 ## Per-platform guides

@@ -50,6 +50,7 @@ and Codex all read.
 | GitHub Copilot | [copilot-setup.md](copilot-setup.md) | `.github/skills`, `.claude/skills`, `~/.copilot/skills` | ✅ |
 | Kiro | [kiro-setup.md](kiro-setup.md) | `.kiro/skills`, `~/.kiro/skills` | ✅ |
 | Factory Droid | [factory-droid-setup.md](factory-droid-setup.md) | `~/.factory/skills`, `.factory/skills`, `.agent/skills` | ✅ |
+| DeepSeek Harness | [dsh-setup.md](dsh-setup.md) | `.dsh/skills`, `.agents/skills`, `~/.dsh/skills`, `~/.agents/skills` | ✅ |

 For Claude Code, Gemini CLI, and Codex CLI, see the [README install section](../README.md#installation).
 `./scripts/install.sh gemini` and `./scripts/install.sh codex` also work and use the flat layout these
@@ -68,7 +69,7 @@ If your agent accepts a skills folder or an instruction file, brooks-lint works:
 ## Verification status

 The marketplace-installed platforms (Claude Code, Gemini CLI, Codex CLI) are maintainer-verified. The
-eight Agent-Skills platforms above are documented from each tool's official skill spec and verified at
+nine Agent-Skills platforms above are documented from each tool's official skill spec and verified at
 the file-layout level (the installer is tested), but not yet end-to-end run by the maintainer on every
 platform. **Tried one? Tell us** — [open an issue](https://github.com/hyhmrright/brooks-lint/issues/new)
 with the platform, version, and what you saw, working or broken.
@@ -28,3 +28,4 @@ jobs:
          # fail-on: critical            # fail on any Critical finding (none | warning | critical)
          # fail-on-regression: true     # fail if the Health Score dropped vs the last run
          # sarif-file: brooks-lint.sarif  # also upload findings to GitHub Code Scanning
+          # api-base-url: https://your-gateway.example.com  # any Anthropic-compatible /v1/messages endpoint
@@ -28,7 +28,7 @@
  "description": "AI code reviews grounded in twelve classic software engineering books. Decay-risk diagnostics with book citations, severity labels, and six analysis modes including full-sweep auto-fix.",
  "url": "https://hyhmrright.github.io/brooks-lint/",
  "image": "https://hyhmrright.github.io/brooks-lint/hero.png",
-  "softwareVersion": "1.4.3",
+  "softwareVersion": "1.5.0",
  "license": "https://github.com/hyhmrright/brooks-lint/blob/main/LICENSE",
  "codeRepository": "https://github.com/hyhmrright/brooks-lint",
  "keywords": "AI code review, code quality, tech debt, architecture audit, test quality, Claude Code plugin, refactoring, clean architecture",
@@ -1,6 +1,6 @@
 {
  "name": "brooks-lint",
-  "version": "1.4.3",
+  "version": "1.5.0",
  "description": "AI code reviews grounded in twelve classic engineering books — decay risk diagnostics with book citations, severity labels, and six analysis modes (PR review, architecture audit, tech debt, test quality, health dashboard, full-sweep auto-fix)",
  "author": "hyhmrright",
  "license": "MIT",
@@ -1,6 +1,6 @@
 {
  "name": "brooks-lint",
-  "version": "1.4.3",
+  "version": "1.5.0",
  "type": "module",
  "scripts": {
    "bump": "node scripts/bump-version.mjs",
@@ -101,7 +101,9 @@ try {
  process.exit(1);
 }

-const report = message.content[0]?.text ?? "";
+// The first block is not always the text one — an endpoint that returns a
+// thinking block first would otherwise yield undefined.
+const report = message.content.find((block) => block.type === "text")?.text ?? "";

 const scoreMatch = report.match(/Health\s+Score[:\s]+(\d+)/i);
 const score = scoreMatch ? parseInt(scoreMatch[1], 10) : null;
@@ -0,0 +1,241 @@
+// Generates the README star-history chart from first-party GitHub data.
+//
+// GitHub restricted the public stargazers API to a repository's own admins and
+// collaborators (announced 2026-06-30), which broke every third-party chart
+// service — api.star-history.com now serves an error card instead of a chart.
+// We own this repo, so we read the star data ourselves and commit the result,
+// keeping the README free of any third-party image host.
+//
+// The committed dataset, not the drawing, is the source of truth: the SVG is a
+// pure function of assets/star-history.json. That keeps the chart re-renderable
+// with no credentials if the endpoint tightens further, keeps the weekly diff
+// readable (dates, not shifted path coordinates), and lets `npm run validate`
+// prove the two files agree.
+//
+// Run: node scripts/gen-star-history.mjs                → refetch, rewrite both files
+//      node scripts/gen-star-history.mjs --render-only   → redraw the SVG offline
+import { execFileSync } from "node:child_process";
+import { readFileSync, writeFileSync } from "node:fs";
+import { fileURLToPath } from "node:url";
+import { dirname, join } from "node:path";
+
+const ROOT = join(dirname(fileURLToPath(import.meta.url)), "..");
+// README-only assets, so they live in assets/ alone — unlike the banners, the
+// GitHub Pages site under docs/ does not render them.
+const OUT = join(ROOT, "assets", "star-history.svg");
+const DATA = join(ROOT, "assets", "star-history.json");
+
+const REPO = process.env.GITHUB_REPOSITORY ?? "hyhmrright/brooks-lint";
+
+const W = 800, H = 400;
+const PAD = { top: 44, right: 24, bottom: 48, left: 64 };
+const PLOT_W = W - PAD.left - PAD.right;
+const PLOT_H = H - PAD.top - PAD.bottom;
+const ACCENT = "#3b82f6"; // matches the logo palette used by gen-banner.mjs
+const FONT = `-apple-system, BlinkMacSystemFont, "Segoe UI", Helvetica, Arial, sans-serif`;
+
+const esc = (s) => String(s).replace(/&/g, "&amp;").replace(/</g, "&lt;").replace(/>/g, "&gt;");
+
+// The stargazers endpoint is no longer public: it needs credentials that can
+// read this repo. CI supplies GITHUB_TOKEN; locally we borrow the gh CLI's.
+function resolveToken() {
+  if (process.env.GITHUB_TOKEN) return process.env.GITHUB_TOKEN;
+  try {
+    return execFileSync("gh", ["auth", "token"], { encoding: "utf8" }).trim();
+  } catch {
+    throw new Error(
+      "No GITHUB_TOKEN set and `gh auth token` failed. Since GitHub restricted " +
+        "the stargazers API, this script needs credentials for a repo admin or collaborator.",
+    );
+  }
+}
+
+// Returns the raw starred_at strings, oldest first — exactly what we commit.
+async function fetchStarTimestamps(token) {
+  const stamps = [];
+  for (let page = 1; ; page++) {
+    const url = `https://api.github.com/repos/${REPO}/stargazers?per_page=100&page=${page}`;
+    const res = await fetch(url, {
+      headers: {
+        // The star+json media type is what adds starred_at to each entry.
+        Accept: "application/vnd.github.star+json",
+        Authorization: `Bearer ${token}`,
+        "X-GitHub-Api-Version": "2022-11-28",
+        "User-Agent": "brooks-lint-gen-star-history",
+      },
+    });
+    // Past 400 pages GitHub answers 422 rather than paginating, so a repo above
+    // 40,000 stars can no longer be read in full — the committed dataset is what
+    // preserves the history when that day comes.
+    if (!res.ok) {
+      throw new Error(`GitHub API ${res.status} on page ${page}: ${(await res.text()).slice(0, 200)}`);
+    }
+    const batch = await res.json();
+    for (const entry of batch) {
+      // A missing starred_at means the star+json media type stopped being
+      // honoured. Fail loudly rather than plot NaN coordinates.
+      if (Number.isNaN(Date.parse(entry.starred_at))) {
+        throw new Error(`Stargazer without a usable starred_at on page ${page}.`);
+      }
+      stamps.push(entry.starred_at);
+    }
+    if (batch.length < 100) return stamps.sort((a, b) => Date.parse(a) - Date.parse(b));
+  }
+}
+
+export function readStamps() {
+  return JSON.parse(readFileSync(DATA, "utf8")).starredAt;
+}
+
+function writeStamps(stamps) {
+  // Deliberately no generated-at field: the file has to stay byte-identical when
+  // no star was added, or the workflow's "commit only when it moved" guard would
+  // fire every single run. Git already records when it last changed.
+  writeFileSync(DATA, `${JSON.stringify({ repo: REPO, starredAt: stamps }, null, 2)}\n`);
+}
+
+// Round the axis maximum up to a 1/2/5 × 10ⁿ step so tick labels stay readable.
+function niceStep(max, targetTicks) {
+  const raw = max / targetTicks;
+  const mag = 10 ** Math.floor(Math.log10(raw));
+  for (const m of [1, 2, 5]) if (raw <= m * mag) return m * mag;
+  return 10 * mag;
+}
+
+// One tick per month, thinned out so labels never collide on a long history.
+function monthTicks(from, to) {
+  const all = [];
+  const cursor = new Date(from);
+  cursor.setUTCDate(1);
+  cursor.setUTCHours(0, 0, 0, 0);
+  if (cursor.getTime() < from) cursor.setUTCMonth(cursor.getUTCMonth() + 1);
+  while (cursor.getTime() <= to) {
+    all.push(cursor.getTime());
+    cursor.setUTCMonth(cursor.getUTCMonth() + 1);
+  }
+  const stride = Math.ceil(all.length / 8) || 1;
+  return all.filter((_, i) => i % stride === 0);
+}
+
+const MONTHS = ["Jan", "Feb", "Mar", "Apr", "May", "Jun", "Jul", "Aug", "Sep", "Oct", "Nov", "Dec"];
+
+function monthLabel(ms, showYear) {
+  const d = new Date(ms);
+  const month = MONTHS[d.getUTCMonth()];
+  return showYear ? `${month} ${d.getUTCFullYear()}` : month;
+}
+
+// 1,392 points would bloat the committed SVG for no visible gain. Keep every
+// nth sample plus the exact final point, so the headline count stays truthful.
+function downsample(points, limit) {
+  if (points.length <= limit) return points;
+  const stride = Math.ceil(points.length / limit);
+  const kept = points.filter((_, i) => i % stride === 0);
+  if (kept.at(-1) !== points.at(-1)) kept.push(points.at(-1));
+  return kept;
+}
+
+export function render(stamps) {
+  const times = stamps.map((iso) => Date.parse(iso));
+  const total = times.length;
+  const t0 = times[0];
+  // The axis ends at the newest star rather than at "now". Anchoring it to the
+  // clock would shift every x coordinate on every run, so the workflow could
+  // never tell a real change from a redraw and would commit noise weekly.
+  const t1 = times.at(-1);
+  // Stars are whole numbers, so never let a tiny repo produce fractional ticks.
+  const step = Math.max(1, niceStep(total, 5));
+  const yMax = Math.ceil(total / step) * step;
+
+  const x = (ms) => PAD.left + ((ms - t0) / (t1 - t0)) * PLOT_W;
+  const y = (n) => PAD.top + PLOT_H - (n / yMax) * PLOT_H;
+
+  const points = downsample(
+    times.map((ms, i) => [ms, i + 1]),
+    300,
+  );
+  const line = points.map(([ms, n], i) => `${i === 0 ? "M" : "L"}${x(ms).toFixed(1)} ${y(n).toFixed(1)}`).join("");
+  const area = `${line}L${x(t1).toFixed(1)} ${y(0).toFixed(1)}L${x(t0).toFixed(1)} ${y(0).toFixed(1)}Z`;
+
+  const yTicks = [];
+  for (let n = 0; n <= yMax; n += step) yTicks.push(n);
+
+  const xTicks = monthTicks(t0, t1);
+  const spansYears = new Date(t0).getUTCFullYear() !== new Date(t1).getUTCFullYear();
+
+  const gridLines = yTicks
+    .map(
+      (n) =>
+        `<line x1="${PAD.left}" y1="${y(n).toFixed(1)}" x2="${PAD.left + PLOT_W}" y2="${y(n).toFixed(1)}" class="grid"/>` +
+        `<text x="${PAD.left - 10}" y="${(y(n) + 4).toFixed(1)}" class="tick" text-anchor="end">${n.toLocaleString("en-US")}</text>`,
+    )
+    .join("\n    ");
+
+  const xLabels = xTicks
+    .map(
+      (ms, i) =>
+        `<text x="${x(ms).toFixed(1)}" y="${PAD.top + PLOT_H + 22}" class="tick" text-anchor="middle">` +
+        `${monthLabel(ms, spansYears || i === 0)}</text>`,
+    )
+    .join("\n    ");
+
+  // Anchoring the axis to the newest star puts the final point exactly on the
+  // right edge, so the callout always hangs back inside the plot.
+  const lastX = PAD.left + PLOT_W;
+  const lastY = y(total);
+
+  return `<svg xmlns="http://www.w3.org/2000/svg" width="${W}" height="${H}" viewBox="0 0 ${W} ${H}" role="img" aria-label="Star history for ${esc(REPO)}: ${total} stars">
+  <style>
+    .bg { fill: #ffffff; }
+    .title { fill: #111827; font: 600 16px ${FONT}; }
+    .sub { fill: #6b7280; font: 400 12px ${FONT}; }
+    .tick { fill: #6b7280; font: 400 11px ${FONT}; }
+    .grid { stroke: #e5e7eb; stroke-width: 1; }
+    .axis { stroke: #d1d5db; stroke-width: 1; }
+    .total { fill: ${ACCENT}; font: 600 13px ${FONT}; }
+    @media (prefers-color-scheme: dark) {
+      .bg { fill: #0d1117; }
+      .title { fill: #e6edf3; }
+      .sub, .tick { fill: #8b949e; }
+      .grid { stroke: #21262d; }
+      .axis { stroke: #30363d; }
+    }
+  </style>
+  <defs>
+    <linearGradient id="fade" x1="0" y1="0" x2="0" y2="1">
+      <stop offset="0%" stop-color="${ACCENT}" stop-opacity="0.28"/>
+      <stop offset="100%" stop-color="${ACCENT}" stop-opacity="0.02"/>
+    </linearGradient>
+  </defs>
+  <rect width="${W}" height="${H}" class="bg"/>
+  <text x="${PAD.left}" y="26" class="title">Star History</text>
+  <text x="${W - PAD.right}" y="26" class="sub" text-anchor="end">${esc(REPO)}</text>
+  <g>
+    ${gridLines}
+  </g>
+  <line x1="${PAD.left}" y1="${PAD.top + PLOT_H}" x2="${PAD.left + PLOT_W}" y2="${PAD.top + PLOT_H}" class="axis"/>
+  <path d="${area}" fill="url(#fade)"/>
+  <path d="${line}" fill="none" stroke="${ACCENT}" stroke-width="2.5" stroke-linejoin="round" stroke-linecap="round"/>
+  <circle cx="${lastX.toFixed(1)}" cy="${lastY.toFixed(1)}" r="4" fill="${ACCENT}"/>
+  <text x="${(lastX - 12).toFixed(1)}" y="${(lastY + 4).toFixed(1)}" class="total" text-anchor="end">${total.toLocaleString("en-US")}</text>
+  <g>
+    ${xLabels}
+  </g>
+</svg>
+`;
+}
+
+async function main() {
+  const renderOnly = process.argv.includes("--render-only");
+  const stamps = renderOnly ? readStamps() : await fetchStarTimestamps(resolveToken());
+  // Two points are the minimum a time axis can span; one would divide by zero
+  // and silently write a chart full of NaN coordinates.
+  if (stamps.length < 2) throw new Error(`Only ${stamps.length} stargazer(s) for ${REPO} — refusing to write a chart.`);
+  if (!renderOnly) writeStamps(stamps);
+  writeFileSync(OUT, render(stamps));
+  console.log(`Wrote ${OUT} — ${stamps.length} stars through ${stamps.at(-1).slice(0, 10)}`);
+}
+
+// Importable so `npm run validate` can re-render from the committed data and
+// prove the SVG is in sync; only a direct run touches the network or disk.
+if (process.argv[1] === fileURLToPath(import.meta.url)) await main();
@@ -13,8 +13,9 @@
 #   ./scripts/install.sh <platform> [--project]
 #   curl -fsSL https://raw.githubusercontent.com/hyhmrright/brooks-lint/main/scripts/install.sh | bash -s -- <platform>
 #
-# Platforms: opencode cursor windsurf antigravity pi kiro copilot droid gemini codex claude agents
-#   agents = the vendor-neutral ~/.agents/skills folder (read by Cursor, Copilot, pi, Gemini, Codex)
+# Platforms: opencode cursor windsurf antigravity pi kiro copilot droid dsh gemini codex claude agents
+#   agents = the vendor-neutral ~/.agents/skills folder (read by Cursor, Copilot, pi, Gemini,
+#            Codex, and DeepSeek Harness)
 #
 # Flags:
 #   --project   install into the current repo (./.<platform>/skills) instead of the global folder
@@ -24,7 +25,7 @@
 set -euo pipefail

 REPO_URL="https://github.com/hyhmrright/brooks-lint.git"
-PLATFORMS="opencode cursor windsurf antigravity pi kiro copilot droid gemini codex claude agents"
+PLATFORMS="opencode cursor windsurf antigravity pi kiro copilot droid dsh gemini codex claude agents"

 err()  { printf '\033[31merror:\033[0m %s\n' "$*" >&2; }
 info() { printf '\033[36m›\033[0m %s\n' "$*"; }
@@ -59,6 +60,8 @@ global_dir() {
    kiro)        printf '%s' "$HOME/.kiro/skills" ;;
    copilot)     printf '%s' "$HOME/.copilot/skills" ;;
    droid)       printf '%s' "$HOME/.factory/skills" ;;
+    # DeepSeek Harness resolves its config root from $DSH_HOME, falling back to ~/.dsh.
+    dsh)         printf '%s' "${DSH_HOME:-$HOME/.dsh}/skills" ;;
    gemini)      printf '%s' "$HOME/.gemini/skills" ;;
    codex)       printf '%s' "$HOME/.codex/skills" ;;
    claude)      printf '%s' "$HOME/.claude/skills" ;;
@@ -77,6 +80,7 @@ project_dir() {
    kiro)        printf '%s' "$PWD/.kiro/skills" ;;
    copilot)     printf '%s' "$PWD/.github/skills" ;;
    droid)       printf '%s' "$PWD/.factory/skills" ;;
+    dsh)         printf '%s' "$PWD/.dsh/skills" ;;
    gemini)      printf '%s' "$PWD/.gemini/skills" ;;
    codex)       printf '%s' "$PWD/.codex/skills" ;;
    claude)      printf '%s' "$PWD/.claude/skills" ;;
@@ -0,0 +1,64 @@
+/**
+ * Platform-inventory helpers shared by validate-repo.mjs and its tests.
+ *
+ * Nothing here hardcodes a platform or a translation. The platforms carrying an
+ * install-table row are discovered from the docs/<name>-setup.md files on disk,
+ * the documents that must show that table are discovered from README*.md, and
+ * the installer's own list is parsed out of scripts/install.sh — so a new
+ * platform, or a seventh language, is covered on arrival. Keeping a separate
+ * hand-maintained list is what let the localized README badges go stale before
+ * (see version-refs.mjs).
+ */
+
+import { readdirSync } from "node:fs";
+import path from "node:path";
+
+/**
+ * Every document that carries the per-platform install table: all README
+ * translations, plus the getting-started guide, which is the one non-README
+ * page with a platform table of its own.
+ */
+export function platformDocs(root) {
+  const readmes = readdirSync(root)
+    .filter((file) => file.startsWith("README") && file.endsWith(".md"))
+    .sort();
+  return [...readmes, path.join("docs", "getting-started.md")];
+}
+
+/** Every per-platform setup guide, as bare filenames. */
+export function setupGuides(root) {
+  return readdirSync(path.join(root, "docs"))
+    .filter((file) => file.endsWith("-setup.md"))
+    .sort();
+}
+
+/**
+ * Setup-guide filenames linked from one document, normalized so a README's
+ * `docs/kiro-setup.md` and getting-started's sibling `kiro-setup.md` compare
+ * equal. Returns a sorted, de-duplicated array.
+ */
+export function linkedSetupGuides(text) {
+  const links = text.matchAll(/\((?:docs\/)?([a-z0-9-]+-setup\.md)\)/g);
+  return [...new Set([...links].map((match) => match[1]))].sort();
+}
+
+/**
+ * The installer's platform list plus the platforms each directory-mapping
+ * function actually handles. A platform in PLATFORMS with no case arm fails
+ * `install.sh <platform>` with "unknown platform"; a case arm missing from
+ * PLATFORMS is invisible in --list and the help text.
+ */
+export function parseInstallerPlatforms(text) {
+  const declared = text.match(/^PLATFORMS="([^"]*)"/m)?.[1].trim();
+  return {
+    declared: declared ? declared.split(/\s+/) : [],
+    global: caseArms(text, "global_dir"),
+    project: caseArms(text, "project_dir"),
+  };
+}
+
+/** Platform names matched by the `case` arms of one shell function. */
+function caseArms(text, fnName) {
+  const body = text.match(new RegExp(`^${fnName}\\(\\) \\{$([\\s\\S]*?)^\\}$`, "m"))?.[1] ?? "";
+  return [...body.matchAll(/^\s+([a-z][a-z0-9-]*)\)/gm)].map((match) => match[1]);
+}
@@ -81,7 +81,7 @@ for (const scenario of scenarios) {
      system: systemPrompt,
      messages: [{ role: "user", content: userMessage }],
    });
-    aiText  = message.content[0]?.text ?? "";
+    aiText  = message.content.find((block) => block.type === "text")?.text ?? "";
    verdict = classify(scenario, aiText);
  } catch (err) {
    error   = err.message;
@@ -15,6 +15,8 @@ import {
 } from "./frontmatter.mjs";
 import { GUIDE_BY_MODE, VALID_MODES } from "./assemble-prompt.mjs";
 import { versionRefs } from "./version-refs.mjs";
+import { platformDocs, setupGuides, linkedSetupGuides, parseInstallerPlatforms } from "./platforms.mjs";
+import { render as renderStarHistory, readStamps as readStarStamps } from "./gen-star-history.mjs";

 const __dirname = path.dirname(fileURLToPath(import.meta.url));
 const root = path.resolve(__dirname, "..");
@@ -325,6 +327,49 @@ function checkAgentsDocs() {
  }
 }

+// Every docs/<name>-setup.md must be linked from all six READMEs and the
+// getting-started table. A platform added to one language and forgotten in the
+// others was previously caught only by hand.
+function checkPlatformDocs() {
+  const guides = setupGuides(root);
+  check(guides.length > 0, "docs/ should contain at least one <platform>-setup.md guide");
+
+  for (const file of platformDocs(root)) {
+    const linked = linkedSetupGuides(readText(file));
+    for (const guide of guides) {
+      check(linked.includes(guide), `${file} is missing an install-table link to docs/${guide}`);
+    }
+    for (const link of linked) {
+      check(guides.includes(link), `${file} links to docs/${link}, which does not exist`);
+    }
+  }
+}
+
+function checkInstallerPlatforms() {
+  const { declared, global: globalArms, project } = parseInstallerPlatforms(readText("scripts/install.sh"));
+  check(declared.length > 0, "scripts/install.sh should declare a PLATFORMS list");
+
+  for (const [arms, fn] of [[globalArms, "global_dir"], [project, "project_dir"]]) {
+    for (const platform of declared) {
+      check(arms.includes(platform), `scripts/install.sh ${fn}() has no path for PLATFORMS entry '${platform}'`);
+    }
+    for (const platform of arms) {
+      check(declared.includes(platform), `scripts/install.sh ${fn}() maps '${platform}', which PLATFORMS omits`);
+    }
+  }
+}
+
+// assets/star-history.svg is a pure function of assets/star-history.json, so a
+// mismatch means the chart was hand-edited or the data moved without a redraw.
+// Re-rendering here needs no credentials, which is the point of committing the
+// data rather than the drawing alone.
+function checkStarHistory() {
+  check(
+    renderStarHistory(readStarStamps()) === readText("assets/star-history.svg"),
+    "assets/star-history.svg does not match assets/star-history.json — rerun `node scripts/gen-star-history.mjs --render-only`",
+  );
+}
+
 function checkSecurity() {
  const security = readText("SECURITY.md");
  check(!security.includes("<!--"), "SECURITY.md still contains placeholder content");
@@ -373,7 +418,10 @@ checkStepAlignment();
 checkEvalSuite();
 checkContributing();
 checkAgentsDocs();
+checkPlatformDocs();
+checkInstallerPlatforms();
 checkSecurity();
+checkStarHistory();
 checkHookOutput();

 // ── Report ─────────────────────────────────────────────────────────────────
@@ -31,6 +31,7 @@ import { reportToSarif } from "./sarif.mjs";
 import { severityBreached, isRegression } from "./ci-gate.mjs";
 import { summarize } from "./benchmark.mjs";
 import { versionRefs } from "./version-refs.mjs";
+import { linkedSetupGuides, parseInstallerPlatforms } from "./platforms.mjs";

 const __dirname = path.dirname(fileURLToPath(import.meta.url));

@@ -905,6 +906,70 @@ test("patterns match the real badge and JSON-LD shapes", () => {
  });
 });

+console.log("\nlinkedSetupGuides");
+
+test("collects guides from README-style and sibling-style links alike", () => {
+  const text = "| Kiro | [setup](docs/kiro-setup.md) |\n| pi | [pi-setup.md](pi-setup.md) |";
+  assert.deepEqual(linkedSetupGuides(text), ["kiro-setup.md", "pi-setup.md"]);
+});
+
+test("de-duplicates a guide linked more than once", () => {
+  const text = "[a](docs/dsh-setup.md) … [b](dsh-setup.md)";
+  assert.deepEqual(linkedSetupGuides(text), ["dsh-setup.md"]);
+});
+
+test("returns an empty array when no guide is linked", () => {
+  assert.deepEqual(linkedSetupGuides("no links here"), []);
+});
+
+console.log("\nparseInstallerPlatforms");
+
+const INSTALLER_FIXTURE = [
+  'PLATFORMS="kiro dsh"',
+  "",
+  "global_dir() {",
+  "  case $1 in",
+  "    kiro)        printf '%s' \"$HOME/.kiro/skills\" ;;",
+  "    # DeepSeek Harness resolves its config root from $DSH_HOME.",
+  "    dsh)         printf '%s' \"${DSH_HOME:-$HOME/.dsh}/skills\" ;;",
+  "    *)           return 1 ;;",
+  "  esac",
+  "}",
+  "",
+  "project_dir() {",
+  "  case $1 in",
+  "    kiro)        printf '%s' \"$PWD/.kiro/skills\" ;;",
+  "    *)           return 1 ;;",
+  "  esac",
+  "}",
+].join("\n");
+
+test("reads the declared list and both directory mappings", () => {
+  const parsed = parseInstallerPlatforms(INSTALLER_FIXTURE);
+  assert.deepEqual(parsed.declared, ["kiro", "dsh"]);
+  assert.deepEqual(parsed.global, ["kiro", "dsh"]);
+});
+
+test("omits a platform whose case arm is missing, so the validator can catch it", () => {
+  // project_dir() in the fixture handles kiro but not dsh — running
+  // `install.sh dsh --project` would die with "unknown platform".
+  assert.deepEqual(parseInstallerPlatforms(INSTALLER_FIXTURE).project, ["kiro"]);
+});
+
+test("skips comments and the catch-all arm", () => {
+  const { global: arms } = parseInstallerPlatforms(INSTALLER_FIXTURE);
+  assert.ok(!arms.includes("*"));
+  assert.equal(arms.length, 2);
+});
+
+test("parses the real installer, proving the patterns still match", () => {
+  const installer = readFileSync(path.join(__dirname, "install.sh"), "utf8");
+  const { declared, global: globalArms, project } = parseInstallerPlatforms(installer);
+  assert.ok(declared.length >= 12, `expected the full platform list, got ${declared.length}`);
+  assert.deepEqual(new Set(globalArms), new Set(declared));
+  assert.deepEqual(new Set(project), new Set(declared));
+});
+
 // ── Integration: validate-repo.mjs passes against current repo ─────────────

 console.log("\nvalidate-repo integration");