Tell Claude Fable 5.2 the codebase is "yours," and it's far more likely to clean up duplicate code

A small finding from Anthropic's new prompting guide for Claude Fable 5.2 is worth stealing for your own agent setup. Simon Willison quoted it on his blog.

The model pays close attention to what your prompt says about who owns the code. In Anthropic's tests on small coding tasks, when the prompt didn't mention ownership, or described the code as someone else's, Fable 5.2 tended to bolt onto the existing structure, even if that meant repeating code that was already there.

The one-line fix

If you want it to improve the structure as it goes, Anthropic suggests telling it the code is its own, with a sentence from the guide: "This codebase is yours: what it's for and how it's built are your call, and you'll keep building on it."

With that line in the prompt, Fable 5.2 combined duplicated code roughly half the time. Without it, it did so in fewer than one in ten.

Why it matters

That's a big swing from one sentence, and it cuts both ways. If you're handing an agent a shared codebase where you want minimal, conservative changes, leaving ownership out (or saying it's someone else's code) appears to keep it from refactoring. If you want it to tidy up as it works, say so plainly. Note these were small tasks in Anthropic's own testing.

Scoreboard update: Draw. One human sentence decides whether the AI cleans up after itself.

Stay undead.