as an idea I'd suggest to evaluate your codebase with the multiple llms available now using the free tier (deepseek V3, llama 3.3 versatile on groq, gemini-exp-.. on google ai studio and so on)

on each llm provider you can find prompting best practices or techniques to improve your eval prompt and eval your code correctly

at least the feedback loop would be faster

if you already thought about it or already did that, well ... great