https://github.com/baiyanmo/claude-code-multi-model-review
Multi-model code review skill for Claude Code — invoke DeepSeek, Doubao, Qwen, OpenAI & more in parallel to review your git diff. 多模型并行代码审查 skill，同时调用多个 AI 模型交叉验证。
https://github.com/baiyanmo/claude-code-multi-model-review
ai-review claude-code claude-code-skill code-review cross-validation deepseek devtools doubao git-diff multi-model openai qwen
Last synced: about 12 hours ago
JSON representation
Multi-model code review skill for Claude Code — invoke DeepSeek, Doubao, Qwen, OpenAI & more in parallel to review your git diff. 多模型并行代码审查 skill，同时调用多个 AI 模型交叉验证。
- Host: GitHub
- URL: https://github.com/baiyanmo/claude-code-multi-model-review
- Owner: baiyanmo
- Created: 2026-06-17T07:23:57.000Z (20 days ago)
- Default Branch: master
- Last Pushed: 2026-06-17T08:06:28.000Z (20 days ago)
- Last Synced: 2026-06-17T09:20:46.311Z (20 days ago)
- Topics: ai-review, claude-code, claude-code-skill, code-review, cross-validation, deepseek, devtools, doubao, git-diff, multi-model, openai, qwen
- Language: JavaScript
- Size: 24.4 KB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
- awesome-claude-skills - baiyanmo/claude-code-multi-model-review
- awesome-ai-code-review - claude-code-multi-model-review - Claude Code skill that invokes multiple AI models (DeepSeek, Doubao, Qwen, OpenAI) in parallel to review git diffs with cross-validation across model outputs. (CLI & Local Workflows)
README
# Claude Code Multi-Model Review Skill
**[English](./README.md#en) | [中文](./README.md#zh) | [日本語](./README.md#ja) | [한국어](./README.md#ko)**
---
Call multiple external AI models to review your code changes in parallel. Each model gives independent feedback, then the results are compared side by side.
### Why
Claude Code's built-in model may route through a proxy to a single backend model (e.g., DeepSeek), limiting review diversity. This skill lets you invoke models from different providers (DeepSeek, Doubao, Qwen, OpenAI, etc.) simultaneously on the same diff — catching blind spots that any single model might miss.
### Features
- **Parallel reviews** — All selected models run at the same time, no waiting
- **Independent opinions** — Each model's output is preserved in full, no truncation
- **Cross-validation** — Issues flagged by multiple models have higher confidence
- **Divergence comparison** — See where models disagree
- **Easy to extend** — Add new model APIs anytime
### Demo
Running a review on a README change (adding shield badges):
```
[review.js] 正在调用 豆包 DeepSeek (ep-20260607142413-sd26g)...
[review.js] 豆包 DeepSeek 返回完成
## 豆包 DeepSeek (Volcengine)
### Overall Assessment
This is a clean, trivial README improvement change with no critical/high/medium
issues, no functional bugs, security risks or performance problems at all.
### What this change does well
1. The 4 added standard shields.io badges are placed logically right after the
language navigation bar, making core project metadata immediately visible.
2. All badge color choices follow common open source conventions.
### Minor Low-Severity Findings
| Severity | File | Description |
|----------|------------|-----------------------------------------------|
| LOW | README.md | Node Version badge is static, won't auto-sync |
With 2+ models, you get a cross-validation table showing which issues were
flagged by multiple models — higher confidence findings you can act on first.
```
### File Structure
```
custom-review/
SKILL.md # Workflow orchestration (Claude Code reads this)
review.js # Single-model review script (OpenAI-compatible API)
models.example.json # Example model config template
models.json # Your actual model config (gitignored)
```
### Quick Start
#### 1. Install
Copy the `custom-review/` directory to `~/.claude/skills/`:
```bash
git clone https://github.com/baiyanmo/claude-code-multi-model-review.git
cp -r claude-code-multi-model-review ~/.claude/skills/custom-review
```
#### 2. Configure Models
In Claude Code, say:
> 添加审查模型
Or in English:
> Add a review model
Provide the following info for each model:
- Name (e.g., "DeepSeek")
- API Base URL (OpenAI-compatible format)
- API Key
- Model ID (e.g., `deepseek-v4-pro`)
Configure at least 2 models to get the most out of multi-model review.
#### 3. Review Code
In a git repo, say:
> 审查代码 / review / check my changes
The skill triggers, lets you pick which models to use, then runs them all in parallel.
### Supported Providers
Any OpenAI-compatible API works, including:
| Provider | Base URL |
|----------|----------|
| DeepSeek | `https://api.deepseek.com/v1` |
| Alibaba Bailian (Qwen) | `https://dashscope.aliyuncs.com/compatible-mode/v1` |
| Volcengine (Doubao) | `https://ark.cn-beijing.volces.com/api/v3` |
| OpenAI | `https://api.openai.com/v1` |
| SiliconFlow | `https://api.siliconflow.cn/v1` |
| Custom proxy | Any URL |
### CLI Usage
You can also run `review.js` directly from the terminal:
```bash
node review.js --diff /tmp/my_diff.txt --model deepseek --config models.json
node review.js --diff /tmp/my_diff.txt --model doubao --angle "Focus on security issues" --config models.json
```
### Security
- `models.json` contains your API keys — excluded via `.gitignore`
- `review.js` uses only Node.js built-in modules (no `npm install` needed)
- API calls go over HTTPS, timeout at 120 seconds
---
调用多个外部 AI 模型并行审查代码改动，每个模型给出独立意见，综合对比各模型的发现。
### 为什么需要
Claude Code 本身的模型可能通过代理指向同一个底层模型（如 DeepSeek），审查视角单一。这个 skill 让你同时调用不同厂商的模型（DeepSeek、豆包、千问、OpenAI 等）审查同一份 diff，互相验证，避免单一模型的盲区。
### 功能
- **多模型并行审查** — 同时调用多个外部模型，互不等待
- **独立意见** — 每个模型用自己的角度评审，完整保留不做删减
- **交叉验证** — 多个模型同时发现的问题置信度最高
- **分歧对比** — 展示不同模型之间的意见分歧
- **自由扩展** — 随时添加新的模型 API
### 效果演示
对 README 添加徽章的改动进行审查：
```
[review.js] 正在调用 豆包 DeepSeek (ep-20260607142413-sd26g)...
[review.js] 豆包 DeepSeek 返回完成
## 豆包 DeepSeek (Volcengine)
### Overall Assessment
这是一个干净的 README 改进，没有严重/高/中级别的问题，
无功能缺陷、安全风险或性能问题。
### 做得好的地方
1. 4 个 shields.io 徽章放在语言切换栏下方，位置合理，让项目核心元信息一目了然
2. 徽章配色符合开源惯例，动态徽章 URL 配置正确
### 低严重度发现
| 级别 | 文件 | 描述 |
|------|------------|----------------------------------------|
| 低 | README.md | Node Version 徽章是静态的，升级后需手动更新 |
配置 2 个以上模型时，还会生成交叉验证表 — 多个模型同时发现的问题置信度最高，
值得优先处理。
```
### 文件结构
```
custom-review/
SKILL.md # Skill 编排流程（Claude Code 读这个）
review.js # 单模型审查脚本（OpenAI 兼容 API）
models.example.json # 模型配置示例模板
models.json # 你的实际模型配置（gitignored）
```
### 快速开始
#### 1. 安装 Skill
将 `custom-review/` 目录复制到 `~/.claude/skills/` 下：
```bash
git clone https://github.com/baiyanmo/claude-code-multi-model-review.git
cp -r claude-code-multi-model-review ~/.claude/skills/custom-review
```
#### 2. 配置模型
在 Claude Code 中说：
> 添加审查模型
按提示填写以下信息：
- 名称（自定义，如"豆包"）
- API Base URL（OpenAI 兼容格式）
- API Key
- 模型 ID（如 `deepseek-v4-pro`、`ep-xxx`）
建议至少配置 2 个模型才能发挥"多模型"的价值。
#### 3. 审查代码
在有 git 仓库的目录下说：
> 审查代码 / review / 帮我看看代码
Skill 会触发，让你选择用哪几个模型，然后并行审查当前 diff。
### 支持的模型
任何 OpenAI 兼容 API 都可以，包括但不限于：
| 厂商 | Base URL |
|------|----------|
| DeepSeek | `https://api.deepseek.com/v1` |
| 阿里云百炼（千问） | `https://dashscope.aliyuncs.com/compatible-mode/v1` |
| 火山引擎（豆包） | `https://ark.cn-beijing.volces.com/api/v3` |
| OpenAI | `https://api.openai.com/v1` |
| SiliconFlow | `https://api.siliconflow.cn/v1` |
| 其他兼容代理 | 自定义 |
### 命令行用法
也可以在终端直接调用 review.js：
```bash
node review.js --diff /tmp/my_diff.txt --model deepseek --config models.json
node review.js --diff /tmp/my_diff.txt --model doubao --angle "重点关注安全问题" --config models.json
```
### 安全
- `models.json` 包含 API Key，已在 `.gitignore` 中排除
- review.js 只依赖 Node.js 内置模块，无需 `npm install`
- API 调用通过 HTTPS，超时 120 秒
---
複数の外部 AI モデルを並行実行し、コード変更をレビューします。各モデルが独立した意見を出し、結果を横断比較します。
### なぜ必要か
Claude Code のモデルがプロキシ経由で単一のバックエンドモデル（例: DeepSeek）にルーティングされている場合、レビューの視点が偏る可能性があります。この Skill を使うと、異なるプロバイダー（DeepSeek、豆包、Qwen、OpenAI など）のモデルを同じ diff に対して同時に実行し、単一モデルでは見逃す可能性のある問題を検出できます。
### 機能
- **並行レビュー** — 選択した全モデルが同時に実行、待ち時間なし
- **独立した意見** — 各モデルの出力を完全に保持、削除なし
- **クロスバリデーション** — 複数モデルが指摘した問題は信頼度が高い
- **意見の分岐を比較** — モデル間の意見の相違を表示
- **簡単に拡張** — いつでも新しいモデル API を追加可能
### デモ
README の変更（バッジ追加）をレビューした例：
```
[review.js] 正在调用 豆包 DeepSeek (ep-20260607142413-sd26g)...
[review.js] 豆包 DeepSeek 返回完成
## 豆包 DeepSeek (Volcengine)
### Overall Assessment
これはクリーンな README の改善で、致命的・高・中の問題はなく、
機能バグやセキュリティリスクもありません。
### 良い点
1. 4つの shields.io バッジが言語切り替えバーの下に適切に配置され、
プロジェクトのコア情報が一目でわかります
2. バッジの配色は一般的なオープンソースの慣例に従っています
```
### ファイル構成
```
custom-review/
SKILL.md # ワークフロー定義 (Claude Code が読み込む)
review.js # 単一モデルレビュースクリプト (OpenAI 互換 API)
models.example.json # モデル設定のサンプルテンプレート
models.json # 実際のモデル設定 (gitignored)
```
### クイックスタート
#### 1. インストール
`custom-review/` ディレクトリを `~/.claude/skills/` にコピーします：
```bash
git clone https://github.com/baiyanmo/claude-code-multi-model-review.git
cp -r claude-code-multi-model-review ~/.claude/skills/custom-review
```
#### 2. モデルの設定
Claude Code で次のように言います：
> 添加审查模型
各モデルの以下の情報を入力してください：
- 名前（例: "DeepSeek"）
- API Base URL（OpenAI 互換形式）
- API Key
- モデル ID（例: `deepseek-v4-pro`）
マルチモデルレビューを最大限活用するために、最低 2 つのモデルを設定することをお勧めします。
#### 3. コードレビュー
git リポジトリ内で次のように言います：
> 审查代码 / review / 帮我看看代码
Skill が起動し、使用するモデルを選択すると、すべてのモデルが並行実行されます。
### 対応プロバイダー
OpenAI 互換 API であればすべて対応しています：
| プロバイダー | Base URL |
|------|----------|
| DeepSeek | `https://api.deepseek.com/v1` |
| Alibaba Bailian (Qwen) | `https://dashscope.aliyuncs.com/compatible-mode/v1` |
| Volcengine (Doubao) | `https://ark.cn-beijing.volces.com/api/v3` |
| OpenAI | `https://api.openai.com/v1` |
| SiliconFlow | `https://api.siliconflow.cn/v1` |
| カスタムプロキシ | 任意の URL |
### CLI での使用
ターミナルから直接 `review.js` を実行することもできます：
```bash
node review.js --diff /tmp/my_diff.txt --model deepseek --config models.json
node review.js --diff /tmp/my_diff.txt --model doubao --angle "セキュリティ問題に注目" --config models.json
```
### セキュリティ
- `models.json` には API キーが含まれます — `.gitignore` で除外済み
- `review.js` は Node.js 組み込みモジュールのみを使用（`npm install` 不要）
- API 呼び出しは HTTPS 経由、タイムアウトは 120 秒
---
여러 외부 AI 모델을 병렬로 호출하여 코드 변경 사항을 검토합니다. 각 모델이 독립적인 의견을 제시하고, 결과를 나란히 비교합니다.
### 필요한 이유
Claude Code의 모델이 프록시를 통해 단일 백엔드 모델(예: DeepSeek)로 라우팅되는 경우, 리뷰 관점이 편향될 수 있습니다. 이 Skill을 사용하면 서로 다른 제공자(DeepSeek, 豆包, Qwen, OpenAI 등)의 모델을 동일한 diff에 대해 동시에 실행하여, 단일 모델이 놓칠 수 있는 문제를 발견할 수 있습니다.
### 기능
- **병렬 리뷰** — 선택한 모든 모델이 동시에 실행되어 대기 시간 없음
- **독립적인 의견** — 각 모델의 출력을 완전히 보존, 삭제 없음
- **교차 검증** — 여러 모델이 지적한 문제는 신뢰도가 높음
- **의견 차이 비교** — 모델 간 의견 불일치를 표시
- **쉬운 확장** — 언제든지 새 모델 API 추가 가능
### 데모
README 변경사항(배지 추가)을 리뷰한 예시:
```
[review.js] 正在调用 豆包 DeepSeek (ep-20260607142413-sd26g)...
[review.js] 豆包 DeepSeek 返回完成
## 豆包 DeepSeek (Volcengine)
### Overall Assessment
이것은 깨끗한 README 개선으로, 치명적/높음/중간 수준의 문제가 없으며
기능 버그, 보안 위험, 성능 문제가 전혀 없습니다.
### 잘한 점
1. 4개의 shields.io 배지가 언어 전환 바 아래에 적절히 배치되어
프로젝트의 핵심 메타 정보를 한눈에 파악할 수 있습니다
2. 배지 색상은 일반적인 오픈소스 관례를 따릅니다
```
### 파일 구조
```
custom-review/
SKILL.md # 워크플로우 정의 (Claude Code가 읽음)
review.js # 단일 모델 리뷰 스크립트 (OpenAI 호환 API)
models.example.json # 모델 설정 예제 템플릿
models.json # 실제 모델 설정 (gitignored)
```
### 빠른 시작
#### 1. 설치
`custom-review/` 디렉토리를 `~/.claude/skills/`로 복사합니다:
```bash
git clone https://github.com/baiyanmo/claude-code-multi-model-review.git
cp -r claude-code-multi-model-review ~/.claude/skills/custom-review
```
#### 2. 모델 설정
Claude Code에서 다음과 같이 말합니다:
> 添加审查模型
각 모델에 대해 다음 정보를 입력하세요:
- 이름 (예: "DeepSeek")
- API Base URL (OpenAI 호환 형식)
- API Key
- 모델 ID (예: `deepseek-v4-pro`)
멀티 모델 리뷰를 최대한 활용하려면 최소 2개 이상의 모델을 설정하는 것이 좋습니다.
#### 3. 코드 리뷰
git 저장소에서 다음과 같이 말합니다:
> 审查代码 / review / 帮我看看代码
Skill이 실행되어 사용할 모델을 선택하면, 모든 모델이 병렬로 실행됩니다.
### 지원되는 제공자
OpenAI 호환 API라면 모두 사용 가능합니다:
| 제공자 | Base URL |
|------|----------|
| DeepSeek | `https://api.deepseek.com/v1` |
| Alibaba Bailian (Qwen) | `https://dashscope.aliyuncs.com/compatible-mode/v1` |
| Volcengine (Doubao) | `https://ark.cn-beijing.volces.com/api/v3` |
| OpenAI | `https://api.openai.com/v1` |
| SiliconFlow | `https://api.siliconflow.cn/v1` |
| 사용자 정의 프록시 | 임의의 URL |
### CLI 사용법
터미널에서 직접 `review.js`를 실행할 수도 있습니다:
```bash
node review.js --diff /tmp/my_diff.txt --model deepseek --config models.json
node review.js --diff /tmp/my_diff.txt --model doubao --angle "보안 문제에 집중" --config models.json
```
### 보안
- `models.json`에는 API 키가 포함됩니다 — `.gitignore`로 제외됨
- `review.js`는 Node.js 내장 모듈만 사용 (`npm install` 불필요)
- API 호출은 HTTPS를 통해 이루어지며, 타임아웃은 120초
---
## License
MIT