pr-reviewer
v1.23.0
Published
CLI tool for automated PR reviews using LLM
Downloads
1,102
Readme
PR Review Tool
A CLI tool for automated code review and PR description generation using LLM (Language Learning Model). This tool analyzes Git diffs, structures the changes, and leverages AI to provide insightful code reviews and PR descriptions.
Features
- 🔍 Automated code review generation
- 📝 PR description generation with diagrams and topics
- 🎯 Multiple review modes (brief, detailed, description)
- 📊 Structured diff analysis
- 🚀 Command-line interface with rich options
- 📋 Comprehensive logging system
- 🛡️ Robust error handling
Prerequisites
- Node.js (v14 or higher)
- npm (v6 or higher)
- Git installed and configured
- OpenAI API key
OpenAI API Key Configuration
Set up your OpenAI API key as an environment variable:
# Linux/macOS
export OPEN_API_KEY=your_openai_api_key
# Windows (Command Prompt)
set OPEN_API_KEY=your_openai_api_key
# Windows (PowerShell)
$env:OPEN_API_KEY="your_openai_api_key"
Installation
- Install the required dependencies:
npm install commander winston
- Install the tool globally:
npm install -g .
Usage
Basic Command Structure
pr-reviewer [options]
Options
Options:
-V, --version output version number
-b, --branch <branch> branch name to review (required)
-m, --mode <mode> review mode (review/description) (default: "review")
-t, --target-branch <branch> target branch to compare against (default: "development")
-v, --verbose enable verbose logging
-o, --output <output> output folder (default: "tmp")
-h, --help display help for command
Examples
# Basic review of a feature branch
pr-reviewer -b feature/new-feature
# Generate detailed description comparing against main branch
pr-reviewer -b feature/new-feature -m description -t main
# Review with verbose logging
pr-reviewer -b feature/new-feature -v
# Show help
pr-reviewer --help
Review Modes
review (default)
- Performs a detailed code review
- Identifies potential issues
- Suggests improvements
- Reviews code style and best practices
description
- Generates comprehensive PR description
- Creates diagrams where applicable
- Lists key changes and impacts
- Provides technical context
brief
- Quick summary of changes
- High-level impact analysis
- Key points for reviewers
Output
The tool generates several output files:
combined.log
: Complete execution logerror.log
: Error-specific logging- Temporary diff files (automatically cleaned up)
- LLM review output (saved to specified location)
How It Works
Initialization
- Parses command-line arguments
- Validates input parameters
- Sets up logging system
Git Operations
- Fetches latest changes
- Checks out specified branch
- Generates diff against target branch
Diff Processing
- Parses Git diff output
- Structures changes by file
- Identifies additions and removals
LLM Integration
- Sends structured changes to LLM
- Processes LLM response
- Generates formatted output
Cleanup
- Removes temporary files
- Logs operation completion
- Handles any errors
Error Handling
The tool includes comprehensive error handling for:
- Git operation failures
- File system issues
- LLM communication problems
- Invalid input parameters
Errors are:
- Logged to error.log
- Displayed in console with context
- Handled gracefully with proper cleanup
Development
Project Structure
pr-reviewer/
├── src/
│ ├── GitOperations.js
│ ├── DiffParser.js
│ ├── PRReviewer.js
│ ├── LLMAdapter.js
│ └── send_to_llm.js
├── logs/
│ ├── combined.log
│ └── error.log
└── package.json
Adding New Features
- Follow the existing class structure
- Implement error handling
- Add appropriate logging
- Update tests if applicable
Troubleshooting
Common issues and solutions:
Git access errors
- Ensure Git is properly configured
- Check repository permissions
API key issues
- Verify OPEN_API_KEY is set
- Check key validity
Parsing errors
- Ensure branch exists
- Check for valid diff output
Customizing Prompts
The tool uses customizable prompts for different review modes. These prompts can be modified by editing the MODE_FOR_PROMPT.js
file in the project root.
Prompt Structure
// prompts.js
module.exports = {
PROMPTS: {
review: `[Your custom review prompt]`,
brief: `[Your custom brief prompt]`,
description: `[Your custom description prompt]`,
},
};
Available Modes
- review: Used for detailed code reviews
- brief: Used for quick summaries
- description: Used for PR descriptions
Customizing Prompts
To modify the prompts:
- Locate
MODE_FOR_PROMPT.js
in the project root - Edit the desired prompt in the
PROMPTS
object - Save the file - changes will take effect immediately
Example Customization
// prompts.js
module.exports = {
PROMPTS: {
review: `
As a senior developer, review this code focusing on:
- Performance implications
- Security concerns
- Best practices
- Potential edge cases
Please be constructive and specific.
`,
brief: `
Provide a 2-3 sentence summary of the changes.
Include:
- Main purpose 🎯
- Key technical changes 🔧
- Impact on users 👥
`,
description: `
Create a comprehensive PR description with:
1. Overview of changes
2. Technical implementation details
3. Testing considerations
4. Visual diagrams where applicable
`,
},
};
Best Practices for Custom Prompts
- Be Specific: Clearly define what you want the LLM to focus on
- Format Consistently: Use clear formatting for better readability
- Include Guidelines: Add specific instructions for output format
- Keep it Focused: Each prompt should have a single clear purpose
- Test Changes: Verify new prompts generate desired outputs
Adding New Modes
To add a new review mode:
Add your new prompt to the
PROMPTS
object:module.exports = { PROMPTS: { // Existing prompts... newMode: `Your new prompt here`, }, };
Use the new mode:
pr-reviewer -b feature/branch -m newMode
License
MIT License - see LICENSE.md for details
Support
For bugs and feature requests, please create an issue in the repository.
Note: This tool is in active development. Please report any issues or suggestions for improvement.
Pushing to NPM
# Increment version
npm version patch|minor|major
# Publish
npm publish