The new CLAUDE.md advice: delete, don’t accumulate
Anthropic stripped more than 80% of Claude Code’s system prompt for models like Opus 5 and Fable 5 and saw no measurable loss on coding evals, according to its context engineering post from July 24, 2026.
Guidance on CLAUDE.md has flipped in seven months. In January 2026, Boris Cherny recommended writing a new rule into the file every time the model got something wrong. After Opus 5 shipped on July 26, he told YC Startup School to delete CLAUDE.md, skills, and hooks every six months and add lines back only when a failure repeats.
The tooling already reflects that. In Claude Code 2.1.206, /doctor suggests cutting directory structure, dependency lists, and architecture overviews out of the file, leaving the gotchas and the deviations from defaults.
Up next
Claude Opus 4.5 squeezed a kernel from 147,734 cycles to 1,487
Claude Opus 4.5 cut a baseline kernel from 147,734 cycles to 1,487 on Anthropic’s newly open-sourced VLIW SIMD performance-engineering task, a 99x speedup that beat most human results. The binding constraint is 2 loads per cycle, not compute.
Related stories
- Coding-agent prompt files grow 226% and never shrink
- Enforce your CLAUDE.md rules in CI, not in a markdown file
- Claude Code Field Kit adds a tested Bash safety hook
- Domain-Driven Design as a way to manage Claude’s context
- Missing Claude’s cache on Opus 5.5 costs up to 40x more
- Claude Code cut over 80% of its system prompt, scores held
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.