| name | anti-slop |
| description | Anti-generic-AI design discipline — bans, tells, and the slop test. Load for ANY new UI surface or visual redesign. Inspired by Impeccable (Apache 2.0). |
Skill: Anti-Slop
The test: if someone could look at the interface and say "AI made that" without doubt, it failed. Committed design choices over safe defaults. Project visual identity from memory-long §conventions + design tokens.
Absolute bans (match-and-refuse — rewrite the element)
- Side-stripe borders (
border-left >1px as colored accent on cards/alerts) → full borders, bg tints, leading icons, or nothing
- Gradient text (
background-clip: text) → solid color; emphasis via weight/size
- Glassmorphism as default → rare and purposeful, or nothing
- Hero-metric template (big number + small label + gradient accent) — SaaS cliché
- Identical card grids (icon + heading + text × N, same size forever)
- Tracked-uppercase eyebrow above every section ("ABOUT" / "PROCESS" kickers) — one deliberate kicker is voice; one per section is AI grammar
- Numbered section scaffolding (01/02/03) by reflex — numbers only when the content IS a real sequence
- Purple-to-blue gradients + ✨ emoji + "Powered by AI" pills — the 2024-25 tell trifecta
- Warm-neutral cream/sand/beige body bg as "warmth" — carry warmth via accent, typography and imagery; body bg is a committed choice
- Nested cards (cards inside cards) → flatten — spacing, typography and dividers, not containers in containers
- Icon tile stacked above every heading (rounded-square icon container over each feature heading — the universal AI card template) → icon beside the heading, or in flow without its own container
- Italic serif display hero (oversized Fraunces/Playfair/Recoleta italic headline) — the universal AI-startup hero; set roman or use a non-serif display face (editorial register may earn it — judge by context)
- Oversized full-sentence H1 — a punchy 1–2 word headline at display size is fine; a long sentence blown up dominates the viewport — set it smaller or tighten the copy
- Monotonous spacing (the same gap everywhere) → tight within groups, generous between sections — rhythm is information
Category-reflex check (two altitudes)
- First-order: can someone guess the palette from the category alone (fintech→navy, health→teal, AI→purple)? That's the training-data reflex — rework.
- Second-order: can they guess your "alternative" from category + anti-reference ("AI tool but not SaaS-cream → editorial-typographic")? The trap one tier deeper — rework until neither is obvious.
Commitment axis (pick a color strategy BEFORE colors)
Restrained (neutrals + one accent ≤10%) · Committed (one saturated color at 30–60%) · Full palette (3–4 named roles) · Drenched (the surface IS the color). Products default restrained; brand pages commit harder. Never drift between strategies mid-project.
Dark vs light is never a default
Write one sentence of physical scene (who uses it, where, under what light, in what mood) — if the sentence doesn't force the answer, it's not concrete enough yet.
Verification
Review any new surface against the bans list explicitly · heading copy tested at every breakpoint (overflow = design bug) · when in doubt, the human decides the identity call — propose 2 committed directions, never 1 safe one.
