Anthropic's Fable 5 guide flips how you prompt it

Hook

The habits that got the most out of older Claude models now leave capability on the table: longer prompts, more rules, tighter leashes. Anthropic said so in writing this week.

The news

On June 9 Anthropic released Claude Fable 5 to the public, its most powerful publicly available model, and shipped a prompting guide for it alongside Claude Mythos 5. The headline for operators is that your old habits undersell it. Anthropic tells you to start at the top of your difficulty range and hand it your hardest problems, because testing it on simple work hides its range. Individual runs now stretch for many minutes and autonomous runs for hours, so it says to check on jobs asynchronously instead of waiting on them. It steers on brief instructions, not enumerated rules; older, over-prescriptive skills "can degrade output quality." Effort is the main dial: high as the default, xhigh for the hardest work, medium or low for routine. The guide is here.

On your desk

Two moves carry the most weight. First, give it a memory file. The guide says store one lesson per file with a one-line summary at the top, record corrections and confirmed approaches alike, and delete the ones that turn out wrong. That is the same persistent-memory pattern this stack already runs. Second, ground its progress: a short instruction to audit every status claim against an actual tool result "nearly eliminated fabricated status reports" in Anthropic's own testing. The old assumption that a longer, more controlling prompt buys better output just expired. Trim the prompt, raise the ceiling on the task.

What you should do next

Open your CLAUDE.md or memory file and add the one-lesson-per-file rule, then read how persistent memory actually works.