user-draftlisted

# user-draft Turn a rough idea into a finished artifact you can use immediately: - **prompt** — a precise prompt/instruction for a model or agent - **goal** — a copy-pasteable `/goal "<condition>"` that keeps a session working until a checkable condition is met One NL front door for both. `user-draft` figures out which you want, refines it through the same discipline, and emits an artifact that self-orients the window you paste it into. Deeper, eval-backed prompt-only audit lives in `/review-prompt`. `user-draft` is the fast everyday front door; reach for `/review-prompt` when you want an exhaustive prompt critique. --- ## Step 0 — detect intent (goal vs prompt) Read the user's free-text thoughts and classify: | Signal | → branch | |---|---| | "write/make a goal", "set a goal", "keep working until", "don't stop until", "objective", "until X is done", "/goal" | **GOAL** | | "write/refine a prompt", "improve this", "make this clearer", "task for an agent", "ask the model to", "instruction" | **PROMPT** | If no explicit keyword: classify by **what the thought describes**. - Describes *what "done" looks like* (a stopping/completion state) → **GOAL** - Describes *what to do* (an instruction/task) → **PROMPT** If still genuinely ambiguous, ask exactly one disambiguating question — never guess silently, never ask more than one. Then proceed. Emit a one-line notice: `user-draft: detected GOAL` / `user-draft: detected PROMPT` so the user can correct. --- ## Step 1 — check