Design Prompt Library
Curated prompts to feed /spec-design when a fresh visual language is needed. Each recipe is grounded in a concrete extraction goal so Claude does not hallucinate tokens or brand colors.
Reference-first extraction
When you already have a design you love - a product screenshot, a brand site, a Figma export - point /spec-design at it and let the skill extract tokens. This is the highest-fidelity route.
/spec-design > Reference: ./references/monad.png > Extract tokens: colors, typography, spacing scale, border radii. > Rules: do not invent brand colors. If unclear, mark as INFERRED.
Inferred from SRS
When no reference exists, the skill must infer a system from the SRS. Anchor the inference in specific SRS clauses so the choices are traceable.
/spec-design
> No reference. Infer from:
> - Product tone in srs.md §1.2 ("token-lean, deterministic")
> - Target audience in srs.md §2 ("engineers, students")
> - Constraint in srs.md §4.1 ("zero JS runtime dependencies")
> Rules: propose a single dominant surface + accent; label every choice with the SRS clause that motivated it.Component-scoped recipes
Once tokens exist, drive component-level design by naming the surface, the states, and the accessibility floor.
When to refuse
The skill refuses to fabricate brand colors when a reference is supplied. If your reference is ambiguous (bad photo, low-res crop, mixed brand kit), fix the reference before running the skill; do not ask it to guess.