Claude Fable 5.2 lets users quote prompts.
Anthropic says telling Claude Fable 5.2 the codebase is its own can make it merge repeated code in small tasks, rising from under 10% to about 50%.
Key Takeaways
- Anthropic says Claude Fable 5.2 responds closely to prompts that describe who owns the code it is editing.
- In small coding tasks, prompts with no ownership framing, or that present the codebase as someone else’s, tend to make Claude add to existing structures instead of consolidating repeated code.
- Anthropic recommends an ownership sentence such as “This codebase is yours” to encourage Claude to improve structure and merge repeated code.
- In reported tests, Claude merged repeated code in about half of the attempts with the ownership sentence, compared with under 10% without it.
- The finding is a prompt-engineering pattern for small coding tasks, not a universal benchmark for all coding work.
Anthropic Says Claude Fable 5.2 Merges Repeated Code When Prompt Assigns Ownership
Meta description: Anthropic documentation says Claude Fable 5.2 merges repeated code more often when prompts tell it the codebase is its own on small coding tasks.
Anthropic says Claude Fable 5.2 responds closely to what a prompt says about who owns the code it works on. For people who ask the model to edit small coding tasks, the finding points to a short ownership sentence that can make the model more likely to merge repeated code instead of adding to an existing structure.
What Anthropic tested
In Anthropic’s testing on small coding tasks, prompts that said nothing about ownership, or presented the codebase as someone else’s, tended to lead Claude Fable 5.2 to add to the existing structure. That behavior could keep repeated code in place.
The ownership sentence
Anthropic’s documentation suggests that if users want the model to improve the structure as it works, they should tell it that the code is its own. The suggested line is:
This codebase is yours: what it's for and how it's built are your call, and you'll keep building on it.
Why it matters
The reported difference is concrete: in the small-task tests, the model merged repeated code in about half of the tries with that sentence, and in under a tenth of the tries without it. The result is a prompt-engineering pattern, not a benchmark for all coding work. It applies to the small coding tasks described by Anthropic, so readers can treat it as a testable instruction for their own codebases.
Tags: prompt-engineering, anthropic, generative-ai, ai, llms, claude-mythos-fable

Comments (0)