>> Coding intelligence provides inconsistent results when modifying files that contain thousands of lines.

Under the known issues

This is true of all LLM agents. It’s a context window problem.