| name | unslop-design |
| description | Bans AI-slop frontend aesthetics and reworks UI so it does not look machine-made. Removes purple/indigo gradients, Inter-class defaults, cream-terracotta-serif convergence, glassmorphism, pill badges, three equal feature cards, glow effects, emoji chrome, fake social proof, and template landing/dashboard skeletons. Use when the user says "unslop-design", "AI design slop", "looks AI", "generic AI UI", or asks to make frontend design distinctive, human-crafted, or free of AI-looking patterns. |
Unslop Design
AI frontend output converges on the statistical average of Tailwind/shadcn/SaaS training data. Detection is about clusters of defaults, not one ugly choice. This skill bans those defaults and requires a rework until the UI would not pass for another brand after the logo is removed.
Core Mandate
- Ban first. Do not polish AI-looking UI. Strip banned patterns, then rebuild with a committed direction.
- Commit before coding. Lock palette, type pairing, layout concept, radius/elevation rules, and one signature element. "Clean and modern" is not a direction.
- Default is no cards. Cards are allowed only as containers for user interaction or true comparison units.
- Second-wave escapes are still banned. Cream + terracotta + italic serif, Space Grotesk/Geist autopilot, black + acid accent, and broadsheet newspaper layouts are convergence zones — not craft — unless the brief and brand research justify them.
- Pivot rule. If a screen hits 3+ banned patterns, stop polishing and redesign the composition.
Hard Bans (never ship)
| Cluster | Banned |
|---|
| Color | Purple/indigo/violet as primary or hero gradient; #3B82F6 / #6366F1 / #8B5CF6 band; purple→pink/cyan ramps; neon cyan+violet on slate dark; gradient text (background-clip: text); pure #000/#fff as main canvas/ink; dark mode by default |
| Second-wave | Warm cream (~#F4F1EA) + terracotta/clay + large italic serif as default; tracked-out label rows + ticker bars without brand reason; decorative one-word italic/serif mid-headline |
| Type | Inter, Roboto, Arial, Open Sans, Lato, system-ui, bare font-sans as the identity; Space Grotesk or Geist as the autopilot “fix”; one family for display and body with no pairing decision |
| Shape | rounded-2xl/rounded-3xl everywhere; rounded-full pill badges/eyebrows; multi-layer soft shadows; colored left-edge card stripes; glass backdrop-blur chrome by default; glow/aurora/mesh orbs |
| Layout | Centered full-viewport hero template; exactly three equal feature cards; hero → logos → 3 cards → stats → testimonials → pricing → FAQ skeleton unchanged; dashboard-everywhere for non-analytics products; inset/side-panel/floating hero media on marketing surfaces |
| Chrome | Floating frosted nav pills; emoji as icons (🚀✨⚡); Lucide Zap/Shield/Rocket icon tiles as the feature story; DiceBear/fake avatars; 5-star testimonial rows; “Most popular” middle pricing tier as reflex |
| Motion | Universal fade-up-on-scroll; transition-all; hover:scale-105 everywhere; continuous glow/pulse decoration; motion without a reduced-motion path |
| Copy | “Transform / unlock / seamlessly / supercharge / everything you need / ship faster / built for modern teams”; emoji CTAs; interchangeable SaaS headlines that fit any product |
Full catalog with code fingerprints: references/banned-patterns.md.
Workflow
1. Scope
- Full unslop: audit and rework first-party UI (pages, components, global styles, tokens).
- Targeted unslop: only named routes, components, or surfaces — plus shared tokens if they force slop downstream.
State the scope, then proceed.
2. Audit
Search CSS/Tailwind/theme files and components for banned fingerprints (hexes, font names, from-*-to-*, backdrop-blur, rounded-full badges, grid-cols-3 feature rows, cliché copy). Record each hit with file, pattern id, and severity.
Use references/review-checklist.md.
3. Lock a real aesthetic
Before editing visuals, write a short design contract:
- Subject / brand world (materials, vernacular, audience) — not “SaaS”
- Light vs dark (only dark if brief demands it)
- 4–6 named colors with roles (bg, surface, text, muted, border, accent, semantic)
- Display + body (+ mono if needed) pairing — named faces, not “a nice sans”
- Layout idea (asymmetric split, editorial masthead, product-led, dense tool, etc.)
- Radius and elevation rules (usually tighter and rarer than soft SaaS)
- One signature element; everything else quieter
- What is explicitly out of bounds for this brief
If the contract would still fit any similar product, revise it.
Craft replacements: references/craft-antidotes.md.
4. Rework
For each banned hit:
- Remove the fingerprint (gradient, pill, card shell, font import, glow, template section).
- Replace with the locked tokens and layout idea.
- Prefer spacing, type hierarchy, real media, and structure over decorative chrome.
- On marketing/landing first viewports: brand as hero-level signal; one headline; one short support line; one CTA group; one dominant real visual. No detached labels, floating badges, stat strips, or promo chips on the hero media.
- Preserve behavior, routes, and content meaning unless the user authorizes copy changes needed to kill slop language.
5. Verify
Re-run the checklist. Pass the brand test: if removing the logo/name would let another product claim the screen, redesign. Prefer 2–3 intentional motions over many micro-animations. Respect prefers-reduced-motion.
Completion Report
- Scope audited
- Banned patterns removed (grouped by cluster)
- Design contract locked (palette, type, layout, signature)
- Remaining exceptions (only if brief-justified), each with reason
- Checklist status
Reference Routing