Telling a capable AI how to do a job now costs you time and quality. What to put in a goal prompt, a before-and-after you can copy, three ways to check the work, and when to keep writing the steps.
A copy of the Fable 5.1 system prompt circulated on release day, and the interesting part is where its length goes: overwhelmingly on how the model should decide rather than what it should say. Seven of those decisions transfer into ordinary prompts as single sentences — including the three most people never write. The source is unverified, and the post explains why that doesn’t matter.
Claude Code now audits your CLAUDE.md for you. Run /doctor prompt-audit to find stale rules, broken references and files that contradict each other, then use the fallback prompt if you’re on an older version.