Currently the LLM must return the COMPLETE file content for every modification. For large files this wastes tokens and money. Add a patch/diff mode where the LLM can return unified diffs instead of full file content. This could reduce token usage by 50-90% for large files with small changes. The code_modifier would need to apply patches using difflib or similar.
Currently the LLM must return the COMPLETE file content for every modification. For large files this wastes tokens and money. Add a patch/diff mode where the LLM can return unified diffs instead of full file content. This could reduce token usage by 50-90% for large files with small changes. The code_modifier would need to apply patches using difflib or similar.