ad198a8501
This tries to short-circuit multiple round-trips to llm for reading code. It's a precursor to trying to context engineer tailored to specific tasks. In initial experiments, it's only marginally faster than regular mode, and burns more tokens.
120 KiB
120 KiB