Frontend Design Deslop

Questions: Ask the user through the environment's question tool — never as plain-text prose. One question at a time, 2–4 tappable options, wait for the answer. If the environment has no question tool, ask in prose with the same options, one at a time.

frontend design deslop

AI-generated UI looks generic for two reasons:

1. No constraints. The model samples the statistical median of 2019-2024 web code — Tailwind UI's bg-indigo-500, Inter, rounded cards, soft shadows. You cannot out-prompt a vacuum.
2. Designing before knowing what you're designing. A corporate landing page, a creative portfolio, a developer-tool landing page, an analytics dashboard, and an ecommerce product page share almost no design DNA — a beautiful aesthetic that fights the artifact's job is its own slop.

The fix is a discipline borrowed from brand design: strategy drives design, never aesthetics first.

1. Commit to words first — what this is, who it serves, the adjectives it must feel like.
2. Translate those words into a typography and color system.
3. Build from tokens.
4. Apply the craft layer (layout, components, motion, iconography, imagery, dark mode, accessibility).
5. Audit.

Target the convergence mechanism, not a frozen blocklist; the slop fingerprint shifts over time (purple gradients in 2022, cream backgrounds and italic-serif heroes in 2026).

This skill does two jobs at once: it de-slops the default AI look, and it designs applications well. A distinctive theme on top of careless components, weak layout, or thoughtless motion still reads as amateur. The mechanisms behind every choice live in references/design-theory.md (hierarchy, Gestalt, CRAP, signal-vs-noise, affordances, the interaction laws); read it once so the rest is reasoning rather than rule-following.

Asking questions (CRITICAL)

Every single user question must go through the question tool (see the Questions: directive above) — never as plain-text prose. The bank is generous; the asking is selective. Do not interrogate.

• Batch related questions; offer 2 to 4 concrete options each.
• Ask only the high-signal subset that changes the design system.
• Infer from context first and confirm inferences rather than re-asking.

Phase 0: Discover and commit to words (do this FIRST, before any code)

Check for an existing DESIGN.md at the project root (and common locations like docs/) before writing any code.

• Exists — read it, honor its tokens, skip the questions it already answers, and extend it rather than starting over.
• Missing — resolve three things below before any pixel. Read references/discovery.md for the full protocol, question bank, and the personality-to-token translation table, and references/artifact-types.md for per-type priorities.

1. WHAT is the artifact? Classify it: marketing/landing page, pricing page, SaaS application, dashboard/data tool, ecommerce, marketplace, mobile app, AI/conversational interface, email/newsletter, blog/editorial publication, onboarding/auth flow, settings/admin/CMS, presentation/deck, docs/API reference, portfolio/brand site, or one of the long-tail types in references/artifact-types.md. Each optimizes for a different thing and has its own layout grammar and density. A composite artifact (a marketing site with an embedded app, an AI chat inside a SaaS app) is designed region by region.
2. WHO and WHY? Audience, positioning (corporate vs creative vs technical vs luxury vs playful), and the single primary action or outcome.
3. Commit to words. Lock 3 to 5 brand adjectives and a 3-word aesthetic essence before any visual exploration. This is the highest-leverage input; it drives type, color, density, radius, and motion. Strategy drives design, never the reverse.

Run discovery adaptively: infer, state inferences, ask the high-signal subset through the question tool, and ground the direction in 1 to 3 references (web-search strong current examples of the exact artifact type and positioning if none are given, then transpose rather than originate). Do not proceed until artifact type, positioning, and the adjectives are locked. On a generic brief (another dashboard, another SaaS landing page) where the category itself has little to say, widen the reference search past the product's own category rather than asking for more creativity within it — see references/divergence.md.

Phase 1: Translate strategy into a design system (the gate)

State these commitments in prose, briefly. Each must follow from Phase 0, not from reflex.

1. Aesthetic commitment. Pick ONE opinionated direction that fits the artifact and the adjectives; generic is the failure mode. See references/aesthetics-library.md. If the user gave a brand or reference, transpose it. Derive 2-3 real candidates before picking one, commit to the winner fully before softening it for legibility, and check the bare structure (not just the styled surface) against the category default — see references/divergence.md for why each of these matters and how to do it without re-deliberating in the same pass.

2. Typography (brand-first). Choose type from personality, not aesthetic preference. Match classification to the adjectives, pick a modular-scale ratio that fits the content, and pair for contrast (display + body) without typographic mud. Never Inter/Roboto/Arial/system as the primary face. See references/typography.md.

3. Color (appropriateness + differentiation). Choose colors for fit with the brand and audience, then find uncontested territory (the indigo/violet band is the red ocean of AI UI; avoid it unless the brief demands it). Build one dominant plus a sharp accent plus neutrals plus semantic states, distributed roughly 60-30-10. Author in OKLCH. See references/color-oklch.md.

4. Token table (emit BEFORE components). Display + body font; type scale (state the ratio and base, 6 steps); spacing base unit; max two radius values; ONE shadow approach (defined edge OR soft elevation, never both on one element); palette with roles (bg, surface, fg, muted, border, accent, accent-fg, success, warning, error) — surface is required, not optional: it is what elevation-by-lightness (see references/dark-mode.md) steps off of. Everything references tokens; no scattered hex/px. Pull a starting set from references/token-sets.md.

5. Signature move. Name the single thing that makes this UI memorable and unmistakably not-default. One per project.

6. Adapter. Pick the stack syntax: plain CSS custom properties, Tailwind v4 @theme, or shadcn semantic tokens. See references/adapters.md. references/token-core.css is the portable source of truth.

Phase 2: Apply the system to the interface (the craft layer)

Tokens make a UI consistent; the craft layer makes it good — this is the "design an application" half of the skill, and the half most AI output skips.

Apply each of the following to the artifact, pulling the matching reference on demand. Density and emphasis vary by artifact type (see references/artifact-types.md); a dashboard applies these very differently from a landing page.

1. Layout and composition. Compose space with intent: a base spacing unit, spacing that is tight within groups and generous between sections, an intentional grid (12-column, modular, or bento where content genuinely varies), at least one brief-specific layout move, and whitespace as a signal of confidence. Break the centered-max-width-column reflex. See references/layout.md.

2. Components and states. Specify every interactive component across its full state matrix (default, hover, active, focus, disabled, loading, error, selected), not just at rest. Get buttons (ranked by importance, not colored by meaning), forms (real labels, correct types, inline validation that keeps input), tables (left-align text, right-align tabular-nums numerals, light separators), navigation, overlays, and the empty/loading/error states right. See references/components.md.

3. Motion. Treat motion as communication, under a duration and easing token scale. Default to ease-out under 300ms, animate only transform and opacity, scale popovers from their trigger, and never animate high-frequency actions. See references/motion.md.

4. Iconography. One grid, one stroke, one radius across the set; do not let the unmodified default starter-kit set define the look. See references/iconography.md.

5. Imagery and illustration. Art-direct imagery as a system. Prefer real product visuals over stock and abstract; avoid the AI/stock fingerprint (people pointing at laptops, gradient blobs, corporate-Memphis, default Midjourney). Use texture and a graphic device to escape flat-slop. See references/imagery.md.

6. Dark mode and theming. If dark mode is in scope, design it (do not invert): near-black not pure black, off-white not pure white, elevation via lightness, desaturated accents, all driven by semantic tokens. See references/dark-mode.md.

7. Accessibility as you build. WCAG 2.2 AA: visible managed focus, keyboard operability, labels, 24px-plus targets, color independence, reduced-motion. Build it in; do not bolt it on. See references/accessibility.md.

At the end of conception, once the direction and craft decisions are locked, suggest to the user a relevant subset of design and component catalogs to mine for concrete ideas and ready implementations, framed as inspiration to transpose through the committed system (never to clone) and with a reminder to verify component licenses. Pick by artifact type and stage rather than dumping the whole list. See references/catalogs.md.

Phase 3: Write DESIGN.md (the durable output)

Everything this skill produces lives in a single DESIGN.md at the project root:

• Discovery context
• Committed aesthetic and signature move
• Typography and color systems
• Tokens
• Spacing/radius/shadow rules
• Craft-layer decisions (layout, components, motion, iconography, imagery, dark mode, accessibility)
• Slop-audit result

Write or update it before or alongside building components, using the schema in references/design-md.md. DESIGN.md is the single source of truth — if the CSS, the adapter, or the components ever drift from it, DESIGN.md wins. On later sessions, Phase 0 reads this file instead of re-running discovery.