review-this is a command-line tool that provides automated code review feedback using Ollama's AI models. It analyzes git changes (either unstaged or between branches) and generates detailed review comments about potential issues, best practices, and improvements.
This is just a fun project I've been working with. To run:
- Analyze unstaged changes or compare branches
- Customizable AI model prompts through Modelfiles
- Generate review reports in both JSON and Markdown formats
- Exclude common configuration files automatically
- Support for reviewing new (untracked) files
- Debug mode for troubleshooting
- Configurable model temperature for response generation
- git
- curl
- jq
- Ollama running locally with the codellama model
- Ensure you have all prerequisites installed
- Clone this repository or download the script
- Make the script executable:
chmod +x review-this
- Place the script in your PATH (optional):
sudo ln -s $(pwd)/review-this /usr/local/bin/
Review unstaged changes in the current repository:
review-thisCompare with a specific branch:
review-this --mode branch --branch main--mode, -m Compare mode: 'current' (unstaged changes) or 'branch' (compare with branch)
--branch, -b Branch to compare against (default: develop)
--model, -ml Ollama model to use (default: codellama)
--temperature, -t Model temperature (default: 0.7, range: 0-2)
--modelfile, -mf Path to custom Modelfile with review prompt
--debug, -d Enable debug output
You can customize the review behavior by creating a Modelfile with your own prompt. This allows you to define specific review criteria, response format, and focus areas.
FROM codellama
SYSTEM """
You are a code reviewer analyzing a git diff. Please evaluate the code for:
- Security vulnerabilities
- Performance issues
- Code quality and best practices
- Potential bugs
- Documentation completeness
Format your response as follows:
1. CRITICAL: List any security or major issues
2. WARNINGS: List performance concerns or potential bugs
3. SUGGESTIONS: List style and documentation improvements
4. If no issues found, respond with "LGTM" (Looks Good To Me)
Each issue should include:
- Issue description
- Location or context
- Recommended fix
"""
To use a custom Modelfile:
review-this --modelfile path/to/your/modelfileThe script generates two types of output files:
review_results.json: Contains the raw review data in JSON formatreview_results.md: A formatted Markdown report of the review findings
[
{
"file": "src/main.js",
"review": "• Missing error handling in async function\n• Unused variable 'config' on line 23",
"timestamp": "2024-02-07T15:30:45Z"
}
]Enable debug mode to get detailed logging:
review-this --debugDebug logs are written to git-review-debug.log in the working directory.
- Only works with text-based files
- Requires Ollama to be running locally
- Large diffs may take longer to process
- Memory usage scales with diff size
- Some models behave differently, so the resulting markdown file might be a bit different. For example,
deepseek-r1has reasoning in the response in between the<think></think>tags. I couldn't figure out a way of omiting those. - I'll update when I see more
Feel free to open issues or submit pull requests for:
- Bug fixes
- New features
- Documentation improvements
- Example Modelfiles