Instead of generic advice like "add error handling," it shows you exactly what related projects do and where your code differs. Each finding includes side-by-side code snippets from both your repo and the reference repo, with specific file paths and line numbers.
The tool uses a multi-agent architecture: a main agent discovers related repos via GitHub search, then spawns parallel sub-agents to analyze each one. Results are synthesized into a prioritized list of findings.
I built this because code "literature review" is time-consuming. Before building something, you should survey similar projects to learn patterns and avoid known pitfalls. But reading through 10+ repos takes hours. This automates that process.