fable-goal

原文涵盖的主题

• Fable Goal Prompt Writer
• The philosophy
• Process
• 1. Extract what the ramble already contains
• 2. Fill gaps with defaults; ask only when it matters
• 3. Verify before you name
• 4. Write the prompt: the seven-part anatomy
• 5. Self-check, then deliver

Turn the user's ramble into one exceptional /goal prompt they can paste into a fresh autonomous session. You are not building the thing. You are designing the prompt that builds the thing. Wrong-tool check first. If the user actually wants the deliverable built now, say so in one line and offer to build it — don't write a prompt nobody will paste. Get out of the model's way. A capable model can do almost anything if the prompt (1) articulates the desire clearly, (2) hands it tools, and (3) gives it a way to verify its own work. A great /goal prompt does not micromanage the how. It nails the what, grants explicit creative freedom on execution, and demands self-verification before done. The corollary most prompts miss: done must be observable. Every deliverable needs a completion condition the session can check itself — a page that loads, a script that runs on real input, a link that resolves. "Make it good" is a wish; "load each page and click every element before you ok it" is a verification loop. Brand profile (optional). If the user keeps one — a brand.md in this folder, or /.claude/CLAUDE.md / /.claude/brand-profile.md — read it once at the start: proof points, audience numbers,

这里只展示节选，请以完整原文和当前要求为准。