`filepaths_string` does not consistently generate correct Python code
Given this prompt:
An example web page using React and HTML. It should have a text box for the user to enter some text with a button to submit it, and on submission another field should be updated to show the text that was put in.
I don't get a parseable Python list when using gpt-3.5-turbo:
By using the same type of refining statements given in
generate_file, I was able to produce a proper response.
yea as a first step we can modify the prompt to offer few shot examples like you did, but longer term (to guarantee it never returns invalid code) i think we may need to pull in some other dependencies to guarantee returning valid code. i'll try to work on that.
I've also had success with using the prompt addendum above:
ONLY return a list of strings, like so:
["index.js", "style.css", "index.html"]
DO NOT include code fences:
```py
[1,2,3]
```