Anthropic's prompting guide for Claude Fable 5.2, quoted by Simon Willison, says the model follows what a prompt claims about code ownership. With no mention of ownership, or with the code framed as someone else's, it tends to extend existing structure even when that repeats code. In Anthropic's small coding tests, saying the codebase was the model's own led to repeated code being merged in about half of tries, against under a tenth otherwise. Teams with agent instruction files may want to test this wording on their own refactoring tasks.