<Skeletonify loading={isLoading}>
<ProfileCard user={user} />
</Skeletonify>You build ProfileCard. Then you build ProfileCardSkeleton. You change the component. The skeleton rots. Every React app has this two trees that say the same thing, forever drifting apart.
Skeletonify kills the twin component.
npm install @inaumanmajeed/skeletonifyimport { Skeletonify } from "@inaumanmajeed/skeletonify";
function Profile({ user, isLoading }) {
return (
<Skeletonify loading={isLoading}>
<ProfileCard user={user} />
</Skeletonify>
);
}Zero config. Works in Next.js, Vite, Remix. SSR-safe.
Skeletonify has three layers. You get all three with zero setup. Each makes your skeletons better.
Reads your JSX and Tailwind classes to generate a skeleton at render time. Pure, synchronous, SSR-safe. No DOM measurement, no hidden renders, no flicker.
Result: a plausible skeleton on the very first load. Good enough for most apps.
After your real UI paints, Skeletonify quietly observes the DOM on idle and caches the true layout. Next time the component loads, the skeleton is pixel-accurate.
Result: skeletons that get smarter every time a user visits. No code changes needed.
Pre-generate descriptors at build time. The skeleton is accurate from the very first paint no learning needed.
npx @inaumanmajeed/skeletonify-generate src/components/ --out .skeletonify --mapimport { registerBuildDescriptors } from "@inaumanmajeed/skeletonify";
import { skeletonMap } from "./.skeletonify/skeletonMap";
// Call once at app startup
registerBuildDescriptors(skeletonMap);Result: pixel-perfect cold-start skeletons. L3 is fully optional if you never run the CLI, L1 and L2 handle everything.
manual fallback > L3 build-time > L2 learned cache > L1 heuristic
Every layer is automatic. Higher layers override lower ones. Nothing breaks if a layer is missing.
- Zero config. Install, wrap, ship.
- Zero flicker. No hidden render, no DOM measurement, no swap.
- SSR-safe. Server and client produce identical output.
- Tailwind-first. Reads
w-*,h-*,rounded-full,flex,gap-*, and more. - Self-improving. L1 on first load, L2 on every load after.
- 5KB gzipped. Zero runtime dependencies.
- Accessible.
aria-busy,role="status", respectsprefers-reduced-motion. - Dark mode. Respects
prefers-color-scheme. - TypeScript-first.
<Skeletonify
loading={boolean} // required skeleton or real UI
fallback?={ReactNode} // optional manual override
id?={string} // optional stable cache key
learn?={boolean} // optional disable L2 for this instance
className?={string} // optional skeleton wrapper class
>
{children}
</Skeletonify>Does it work with Next.js App Router?
Yes. Skeletonify is a client component ('use client'). Inference is deterministic, so server and client render the same skeleton.
Does it need Tailwind? No, but it works best with Tailwind. Without className hints, L1 falls back to element-type defaults. L2 learns the real layout regardless.
What about charts / canvas / iframes?
Opaque to inference. Use the fallback prop:
<Skeletonify loading={isLoading} fallback={<MyChartSkeleton />}>
<Chart />
</Skeletonify>What if I redesign a component? L1 adapts instantly (same JSX). L2 re-learns on the next visit. L3 re-generates when you re-run the CLI. Nothing goes stale automatically.
Is L3 required? No. L1 + L2 cover most use cases. L3 is for teams that want deterministic cold-start accuracy.
git clone https://github.com/inaumanmajeed/Skeletonify.git
cd Skeletonify
npm install
npm test # 62 tests
npm run build # ESM + CJS
npm run demo:dev # interactive playgroundMIT
If Skeletonify saved you an afternoon, star the repo.