This is nice work, and it is great to see the effort taken to show the pipeline so this will work for others. One further extension might be to fine-tune to sp

This is nice work, and it is great to see the effort taken to show the pipeline so this will work for others. One further extension might be to fine-tune to specifically encourage behavior for a client like godot-copilot. I bet you could teach it to obey your particular prompt structure (eg, matching indentation, or inserting code at the right spot without adding it elsewhere). That would really complete the story to make this very usable by everyday people who dont know/care about LLM internals and fine tunings.