gpt-taste

Request a landing page and the skill outputs a design plan followed by React code using @gsap/react and ScrollTrigger. It implements scroll pinning, scrubbing text reveals, image scale effects, and card stacking based on a randomized selection of motion paradigms.

The skill enforces ultra-wide containers like max-w-5xl or max-w-6xl and font sizes using clamp(3rem, 5vw, 5.5rem) so the H1 flows horizontally in 2-3 lines. Narrow containers that cause 4-6 line wraps are treated as a failure.

grid-flow-dense is a Tailwind utility for grid-auto-flow: dense that backfills empty grid cells. The skill requires it on every bento grid and verifies col-span and row-span values interlock so no dead cells or empty corners remain.

Yes, the generated code uses Tailwind utility classes and React components compatible with @gsap/react, fitting standard Next.js or React projects. It also references @phosphor-icons/react for icon marquees.

LLMs default to the first layout option repeatedly, producing identical pages. The simulated Python RNG uses a deterministic seed from the prompt's character count to force varied hero layouts, font stacks, components, and animations each time.

Banned elements include meta-labels like SECTION 01 or QUESTION 05, floating stamp badges on hero text, pill-tags under heroes, raw stats in heroes, emojis, invisible low-contrast button text, and the Inter font.