Inspired by this tweet by bio_bootloader, I wrote a Python script that uses the GPT-4 API to "heal" other pytho scripts. As of now, it catches some basic errors, and in my incredibly limited testing, it fixes them. It now also can write a small program with a prompt passed in and then iteratively heal that program until it works.
You just need an OpenAI API Key in your env and GPT-4 access, although this probably could work well enough with GPT-3.5-turbo. I haven't tested that yet. Then just run it as follows:
python senkovi.py buggy_program.py 0 "optional instructions" if you want to fix bugs. This works pretty decently.
python senkovi.py buggy_program.py 1 "edits you want made to the code" if you want to add a feature or change a feature in the code. I haven't tested this much.
To write a program from scratch, use fabian.py.
python fabian.py "A python script to plot a quadratic equation and save it to an image"
Notice that you do not actually provide a filename, this is probably an oversight, but fabian.py comes up with the filename itself. Running fabian.py will run the program fabian writes as well, be careful! Also, I've found that it has a really hard time with indentation so if you pass in a reminder in the instructions you'll probably end up with better results!
senkovi.py does not catch logic errors unless they crash the programfabian.py has all the problems senkovi.py does, although it seems to write programs correctly on the first tryThe test buggy_program in this repo is from Wolverine.
I would love your contributions. Just make a Pull Request and add what you think might be cool.
senkovi.py named Senkovi after Disra Senkovi from Children of Ruin.fabian.py named Fabian after some of the spiders from Children of Time.