AI Code Review 中文

AI-powered git pre-commit hook for automated code review with customizable rules.

Features

• 🤖 Multi-Model Support: Works with OpenAI、Deepseek and Ollama、LM Studio
• 🔍 Smart Diff Analysis: Focuses on meaningful changes, ignores deletions
• ⚙️ Customizable Rules: Security, performance, style checks
• ✏️ Custom Prompts: Fully customize review criteria and prompts
• 📊 Graded Feedback: High/Medium/Low severity classification
• 🛠 Easy Integration: Simple npm install and config

Installation

npm install --save-dev ai-pre-commit-reviewer

Add to git pre-commit hook:

npx add-ai-review

Or if you have husky installed:

.package.json
{
  "husky": {
    "hooks": {
      "pre-commit": "npx ai-code-review-run"
    }
  }
}

Configuration

Option 1: package.json

"aiCheckConfig": {
  "providerType": "openai",
  "apiKey": "your-api-key",
  "model": "gpt-4",
  "baseURL": "https://api.openai.com/v1",
  "maxChunkSize": 4000,
  "language": "english",
  "checkSecurity": true,
  "checkPerformance": true,
  "checkStyle": false,
  "enabledFileExtensions": ".html, .js, .jsx, .ts, .tsx, .vue"
}

Option 2: .env file

providerType=openai
baseURL=http://localhost:11434
model=gpt-4
maxChunkSize=4000
language=chinese

Full Configuration Options

| Parameter | Type | Default | Description | |-----------|------|---------|-------------| | providerType | string | "openai" | AI provider type (openai、deepseek or ollama、LMStudio) | | apiKey | string | - | Provider API key (not required for Ollama or LMStudio) | | model | string | "gpt-3.5-turbo" | Model name | | temperature | number | 0.2 | Controls randomness of AI output (higher = more random) | | baseURL | string | "https://api.openai.com" (OpenAI)"https://api.deepseek.com" (Deepseek)"http://localhost:11434" (Ollama)"http://127.0.0.1:1234" (LM Studio)| API base URL | | maxChunkSize | number | 12000 | Max diff chunk size (characters) | | customPrompts | string | '' |Custom prompt templates. When provided, these will completely replace the default security (checkSecurity), performance (checkPerformance) and style (checkStyle) checks. | | language | string | "chinese" | Output language | | strict | boolean | true | Fail on API errors | | correctedResult | boolean | true | When the result field in the AI's returned result does not match the specific detection item result in the list, the system will automatically correct the final determination result based on the actual detection problem. | | showNormal | boolean | false | Show low/medium severity issues | | checkSecurity | boolean | true | Enable security checks | | checkPerformance | boolean | true | Enable performance checks | | checkStyle | boolean | false | Enable style checks | | enabledFileExtensions | string | '.html, .js, .jsx, .ts, .tsx, .vue' | File types to review |

Review Process

1. Diff Extraction: Gets staged changes via git diff --cached
2. File Filtering: Only processes specified file extensions
3. Chunk Splitting: Splits large diffs into manageable chunks
4. AI Analysis: Sends chunks to configured AI provider
5. Result Aggregation: Combines results from all chunks
6. Output: Displays issues grouped by severity

Example Output

Find 1 changed files...
Running code review with AI: The content will be reviewed in 1 sessions for better accuracy.

X Code review was not passed.Please fix the following high-level issues and try again.
- src/auth.js: [高] - 安全问题 - 硬编码的API密钥
  Suggested fix: 使用环境变量存储敏感信息
- src/db.js: [中] - 性能问题 - 缺少数据库连接池
  Suggested fix: 实现连接池减少连接开销

√ Code review passed.

Supported Providers

OpenAI

• Required: apiKey
• Optional: model (default: gpt-3.5-turbo), baseURL (default: https://api.openai.com)
• Models: gpt-4, gpt-3.5-turbo

Deepseek

• Required: apiKey
• Optional: model (default: deepseek-chat), baseURL (default: https://api.deepseek.com)
• Models: deepseek-chat、deepseek-reasoner

Ollama (Local AI Models)

• Required: None (runs locally)
• Optional: model (default: gpt-3.5-turbo), baseURL (default: http://localhost:11434)
• Setup:
  1. Install Ollama: https://ollama.ai/
  2. Download models: ollama pull <model-name>
  3. Common models: llama2, codellama, mistral
• Example .env configuration:

  providerType=ollama
  model=codellama
  baseURL=http://localhost:11434

LMStudio (Local AI Models)

• Required: None (runs locally)
• Optional: model (default: qwen/qwq-32b), baseURL (default: http://127.0.0.1:1234)

Troubleshooting

Hook not running

• Verify .git/hooks/pre-commit exists and is executable
• Check file contains node path/to/ai-review.js

API Errors

• Verify API key and base URL (not required for Ollama)
• Check network connectivity
• Set strict: false to allow commit on API errors

No changes found

• Check enabledFileExtensions matches your file types
• Verify changes are staged (git add)

The returned data format does not conform to the specification

The model fails to produce valid JSON output, indicating potential limitations in its instruction-following capability. Ensure the model is capable of generating structured data.

License

ISC