AI Signal 505
CLAUDE.md updates rejected in favor of in-chat corrections
↗
A developer argues that maintaining a CLAUDE.md file for Claude Code becomes a permanent record of grievances and outdated rules, and instead relies on in-chat corrections that expire with the session.
For engineers who use Claude Code, this challenges the common advice to document every recurring issue in CLAUDE.md. The author suggests that such rules can become counterproductive as the model improves, and that in-context corrections may be more effective. It raises a practical question about how to manage AI assistant behavior without accumulating harmful instructions.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
The author refuses to update CLAUDE.md, calling it a 'grievance archive' of past frustrations.
Rules written in anger become permanent and may actively harm a newer model's performance.
The author prefers in-chat corrections, which die with the session, over permanent written rules.
THE CLUSTER