Anthropic just told everyone building agents that most of their prompt engineering is now dead weight. On the day Opus 5 shipped, the Claude Code team stripped over 80% of Claude Code’s system prompt — and coding eval scores didn’t move.
What actually changed
This isn’t a product, it’s a guidance shift with a tool attached. The old playbook was “add more instructions to cover edge cases.” Claude 5 (Opus 5, Fable 5, Sonnet 5) flips that. Those extra rules were guardrails written for dumber models — now they conflict with each other, override the model’s own judgment, and burn tokens for nothing. Anthropic calls the fix unhobbling: delete the crutches, don’t add better ones.
The /doctor command
The best practices ship as claude doctor. Run /doctor inside Claude Code and it audits your CLAUDE.md and skills, flags bloat, and proposes cuts you approve one by one. It’s a diagnostic that puts your context on a diet.
The uncomfortable part: every CLAUDE.md you tuned for older Claude may now be actively hurting you. HN put it at 415 points for a reason.
You Might Also Like
- 27 Agents 109 Skills 88k Github Stars is Everything Claude Code Genius or Over Engineering
- Pi Mono 29k Stars and a 200 Token System Prompt That Rivals Claude Code
- Claude Opus 4 7 Goes Wall Street First 64 4 on Vals Finance 1m Context in Claude Code
- Claude Code is now Running on bun Rewritten in Rust Anthropic Shipped it After the Marketing Engineering Fight
- Claude Context Mode Might be the Best Thing Thats Happened to my Claude Code Sessions
Top AI Product 
Leave a comment