Intel / article

Quoting Prompting Claude Fable 5.2

Source: simonwillison.net
Date: Published Nov 10, 2026

Context

Anthropic's prompting guide for Claude Fable 5.2 says the model takes strong cues from what a prompt says about who owns the code it is working on. In Anthropic's testing on small coding tasks, when a prompt said nothing about ownership or presented the code as someone else's, the model tended to add to the existing structure even when that meant repeating code, and to report nearby problems rather than fix them. Simon Willison quoted the start of that passage along with the one-sentence fix Anthropic suggests: tell the model the codebase is its own, that what it is for and how it is built are its call, and that it will keep building on it.

With that sentence, the model merged repeated code in about half of the tries; without it, in under a tenth. The guide adds that the rate varied by task, that the effect held when the sentence sat inside a long , and that the requested feature worked in nearly every try either way.

The full guide, beyond the excerpt Willison quoted, lists what the sentence costs. The model used about 1.5 times as many output and made several changes nobody asked for, many of them new tests and documentation. Apart from bug fixes, it changed how an existing feature behaves in one try in six to ten, more often at higher effort, and its closing summary often left out small changes, so Anthropic advises reviewing the diff and not only the summary. Anthropic says to leave the sentence out where the user owns the code and wants as few changes as possible, and notes that a scope rule elsewhere in the prompt does not cancel it: with a rule against drive-by refactors earlier in the same prompt, the model still merged repeated code in about a third of the tries. All of these figures come from Anthropic's own testing.

Key takeaways · AI-distilled

Anthropic's prompting guide says Claude Fable 5.2 follows what a prompt implies about who owns the code: when ownership is unstated or the code is framed as someone else's, it tends to add to the existing structure, even when that repeats code.
To get structural cleanup as it works, the guide recommends telling the model the code is its own, for example: "This codebase is yours: what it's for and how it's built are your call, and you'll keep building on it."
In Anthropic's tests on small coding tasks, Fable 5.2 merged repeated code in about half of attempts when the prompt included that ownership sentence, and in under a tenth of attempts without it.

Terms in this piece · Glossary

system prompt — The standing instructions a model receives before any user input — defining its role, rules, tools, and tone for the whole conversation.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.

Read the source simonwillison.net