GPTCodeQualitySharp is a language-agnostic code quality assessment tool that utilizes OpenAI GPT (default=GPT-3.5-Turbo, TODO: GPT4) or compatible text-completion models to analyze and score source code based on multiple conventions, via a large prompt.
This project aims to help developers pin-point areas with poor code quality and improve their projects.
Code is split into workable (truncated) chunks, and analyzed chunk-by-chunk - multiple pass-throughs using different chunk sizes (non-factors of each other) can allow the differential, or average, to be used to identify the precise areas, which avoids prompting for exact lines to flag.
(This is only on GitHub to get it out to the public faster, it's not final)
Original Author: Hunter Watson
Original Repo: https://github.com/hwvs/GPTCodeQualitySharp
🏆 If you make any positive changes, please submit a pull request!
- C#/.NET (CSharp)
High Accuracy - PHP
High Accuracy - Ruby
High Accuracy - PowerShell
High Accuracy - JavaScript
Moderate Accuracy - VB.NET
Moderate Accuracy - HTML/CSS
Low Accuracy
(Disclaimer: Accuracy is fairly subjective, and is compared to my own human opinion. Good code was tested from popular and well-engineered GitHub repositories. Bad code came from random searches for bad code, some of my own code, and by modifying some of the good code to add code-smells such as long chains of useless if-statements. )
- Analyzes source code files and scores them based on multiple conventions
- Stores API calls in an SQLite database to prevent duplicate calls
- Uses multiple pass-throughs of the code to accurately pinpoint poor areas
- Includes a demo application to showcase the API
- Improve the scoring algorithm for better accuracy
- Refactor the codebase, using itself to see how it performs and to fine-tune the prompt
- Add more prompts