As mentioned in the readme, this was directly inspired by the this project:
<a href="https://github.com/mpoon/gpt-repository-loader">https://github.com/mpoon/gpt-repository-loader</a>
And by the top comment on the HN thread about it:
<a href="https://news.ycombinator.com/item?id=35191303" rel="nofollow">https://news.ycombinator.com/item?id=35191303</a>
Thanks to both the creator and commenter there for the idea.<p>This adds another "twist", in that if you run the "take the wheel" version, it'll automatically implement the changes (i.e. overwrite a file with whatever GPT spits out) and show you the git diff. This is obviously potentially unsafe - you're overwriting your files with whatever the LLM spits out, so please be careful and don't run it somewhere with important files, or in a repo with files you wouldn't paste into the chatGPT input box. I wouldn't even install this on a work computer.<p>This is pretty basic and isn't necessarily all that useful for real work, the workflow is make request -> send file(s) -> GPT changes one file. There's no way to continue a conversation, or ask it to make multi-file changes etc. It's fun to play with though!
This is cool! I was going to build something similar this weekend, but you beat me to it.<p>I wonder if, in the short term at least, this will affect how we modularize code. I've already started thinking how, for a few instances at least, I'm going to vertically combine the three parts of MVC, so that GPT would have all the needed context for certain requests.