Working with Claude Fable 5 drives home that your prompts and skills are only a map; the real work happens in the codebase and its constraints. Every time Claude hits an unknown, it guesses what you want—and those guesses can slow you down. The trick is spotting four types of unknowns before Claude makes wrong turns: known knowns (what you’ve told it), known unknowns (gaps you recognize), unknown knowns (things so obvious you never mention them), and unknown unknowns (questions you didn’t even think to ask). To uncover those gaps, the author uses several routines. First is a “blind spot pass,” where you ask Claude to list unknown unknowns based on your background and project context. Next comes brainstorming and prototyping—Claude sketches multiple HTML mock-ups or features so you can pinpoint design or spec assumptions early. Then you run interviews: Claude fires one question at a time to surface ambiguities that might change your architecture. When you know what you want but can’t describe it, you point Fable at existing code—say, a Rust crate implementing backoff—and have it reimplement those semantics in your environment. Once you’ve narrowed your unknowns, ask Claude for a step-by-step implementation plan that highlights likely change points. That plan keeps you aligned on scope, and gives you a safety net for pivots. Over time, sharpening these techniques turns vague prompts into clear directives, helping Claude produce code that actually fits your real-world constraints.