The user wants a throwaway prototype: something to react to before any real code. It settles a look, feel, or flow question that prose cannot.

Fast path: a fork about flow, not look (which step comes first, what a state shows), gets labeled wireframe boxes: no design source, no look rules, no reuse of existing markup, and no DESIGN.md write.

Ground it in a design system before building; never prototype on model defaults, which read as slop. Use the project's DESIGN.md or existing design tokens when they exist. When none exists, get one from the user first: a DESIGN.md, a reference site to match, or a source like fontpairs.co for the type pairing. Do not start until type, color, and spacing trace to a real source.
When that source is a reference site or a type pairing rather than an existing DESIGN.md, write the resolved tokens (type, color, spacing) to the project's DESIGN.md before building; it is the one thing the prototype leaves behind, and rubric reads it.
Tokens leave most of the look open. Build to look.md: the defaults generated UI drifts to, and what replaces them.
Build it with the /design skill, a design canvas the user can see and refine. If /design is not available in this environment, write a single self-contained HTML artifact instead, with the same intent. Stop when the user has reacted.

Rules:

Visual only. Mock the surface with fake data. Wire no backend, keep no state. No production code, no branch, no persistence beyond DESIGN.md. The prototype is the whole deliverable.
Say less: short labels, no helper text or explanatory copy the design does not call for. Generated UI over-explains; a prototype that does is already wrong.
When the direction is open, put a few genuinely different layouts on the canvas to react to, not one polished guess.
Reference what exists: point /design at the real components or pages to match, rather than describing them; in the HTML fallback, read their source and reuse its markup and styles.
It is throwaway. It settles the direction; discard it once the user picks, and carry the decision into the real work.