Incorrect output format from LLM's

Hi,
I’m not sure if that’s a bug of the tool of it ias something to do with the particular LLM’s (or maybe system messages sent to them) as I have following issues:

  • while using Google’s API, by default, I get the code with the path flag which did not belong to the code & the code has incorrect indentation (starts with 3 tabs instead of starting from the begining of the line) → very strange and I decided to correct the files manually instead of ‘playing’ with llm to get it right (not always working so it;s waste of tokens)
  • Anthropic Claude 3.5 → sometimes I saw it was creating some content under home directory and no files were visible/accessible afterwards. Other times, it is not creating whole files but adding parts of the code (maybe it’s a feature, but if so, I’d prefer to have the code in chat for which I need to request additionally
    These were jusrt a few examples - I had issues with Deepseek, but I forgot what they were (if I’m not mistaken, I was getting language identifiers like ‘’'python etc). Still, I’d like to double check if it’s only me or if maybe system prompts might be adjusted? I could try to play with them but if those are common issues, maybe the prompts could be modified from your side?