[Bug]: unnecessary steps, forget existing files, overwrites files

Version

VisualStudio Code extension

Operating System

Ubuntu Linux

What happened?

Overall:

It overwrite files, loosing data on iterations
Human made changes gets lost on iterations
Polluting the folder with existing methods, etc
Dozens of unnecessary steps, and they cannot be jumped
Output text, then output JSON.. double generation/half-cut context as result

My suggestions are:

Before modifying a file, retrieve it and ask to adapt the existing file
Prompting: exclusive JSON output, avoid text to JSON duplication
Keep a permanent state of the filetree, methods, functions ,inputs, outputs within the context as a permanent element.
Chain GPT3.5 to provide basics like "shall we refector?, shall we XXXX ?" so it can be used to cheap & quick data binary classification kind of.

Since the code itself doesn't have a logic per-se and it leverages on agents to operate, it should have different tiers where use-case/purpose can be scoped (ex: QuestionAnswering: gpt3.5-turbo, JSONtoNL: gpt3.5-turb, troubleshoot: gpt4-turbo, etc). If we can integrate this with API GPTs/Agents/Langchain.. so the interpreter/sandbox/etc can be fully leveraged... it may take him 1 hour to make a CRUD.. but, he may actually do it 🍡

Also a dialog where u can just intercede, so u can state points like "lets go directly to the code and assume the libraries are installed"..

Overall, its a bruteforcer.. but at least it can deliver "something".. its also important to have better integration with other free-models.. its not like u wanna spend money on create a python env or pip install a package..

Pythagora-io / gpt-pilot