Claude leaves essay-length comments in code — here's how to redirect them to a side file

Birchlabs · x · 2026-08-25

Birchlabs observes that Claude tends to write long comments in code: these are effectively its CoT sequences, or "tombstone" comments explaining diffs during code review — which likely helps the model more than the human reader.

The suggested fix: let it write, but elsewhere. Keep code comments minimal pointers like "guard for obscure case, see service.annex.md for rationale", moving lengthy reasoning into a separate annex file — preserving the model's reasoning benefits without burying the code.

Original post →

More from coding & agent

coding & agent channel →