Core Web Vitals logo

Core Web Vitals

Organization
NVlabs
core-web-vitals

Optimize Core Web Vitals (LCP, INP, CLS) for better page experience and search ranking. Use when asked to "improve Core Web Vitals", "fix LCP", "reduce CLS", "optimize INP", "page experience optimization", or "fix layout shifts".

Overview

PublisherNVlabs
RepositorySkill2Env
Skill namecore-web-vitals
Stars
189
Forks
18
Bundled files
1
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 1 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by NVlabs on GitHub. Read the source before you install it.

Installation

Install the Core Web Vitals AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/NVlabs/Skill2Env/tree/main/SkillHub/skills/addy-osmani/core-web-vitals
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/NVlabs/Skill2Env.git /tmp/Skill2Env
mkdir -p .claude/skills
cp -r /tmp/Skill2Env/SkillHub/skills/addy-osmani/core-web-vitals .claude/skills/core-web-vitals
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Core Web Vitals in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Core Web Vitals on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Core Web Vitals is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Core Web Vitals optimization

Targeted optimization for the three Core Web Vitals metrics that affect Google Search ranking and user experience.

The three metrics

MetricMeasuresGoodNeeds workPoor
LCPLoading≤ 2.5s2.5s – 4s> 4s
INPInteractivity≤ 200ms200ms – 500ms> 500ms
CLSVisual Stability≤ 0.10.1 – 0.25> 0.25

Google measures at the 75th percentile — 75% of page visits must meet "Good" thresholds.


LCP: Largest Contentful Paint

LCP measures when the largest visible content element renders. Usually this is:

  • Hero image or video
  • Large text block
  • Background image
  • <svg> element

Common LCP issues

1. Slow server response (TTFB > 800ms)

Fix: CDN, caching, optimized backend, edge rendering

2. Render-blocking resources

html
<!-- ❌ Blocks rendering -->
<link rel="stylesheet" href="/all-styles.css">

<!-- ✅ Critical CSS inlined, rest deferred -->
<style>/* Critical above-fold CSS */</style>
<link rel="preload" href="/styles.css" as="style" 
      onload="this.onload=null;this.rel='stylesheet'">

3. Slow resource load times

html
<!-- ❌ No hints, discovered late -->
<img src="/hero.jpg" alt="Hero">

<!-- ✅ Preloaded with high priority -->
<link rel="preload" href="/hero.webp" as="image" fetchpriority="high">
<img src="/hero.webp" alt="Hero" fetchpriority="high">

4. Client-side rendering delays

javascript
// ❌ Content loads after JavaScript
useEffect(() => {
  fetch('/api/hero-text').then(r => r.json()).then(setHeroText);
}, []);

// ✅ Server-side or static rendering
// Use SSR, SSG, or streaming to send HTML with content
export async function getServerSideProps() {
  const heroText = await fetchHeroText();
  return { props: { heroText } };
}

5. Make navigations instant with the Speculation Rules API

For most sites, the LCP a user actually experiences is dominated by the next page they navigate to, not the one they landed on. Telling the browser to prerender likely-next pages on hover collapses that LCP to ~0ms.

html
<script type="speculationrules">
{
  "prerender": [{
    "where": { "href_matches": "/*" },
    "eagerness": "moderate"
  }]
}
</script>

eagerness settings (cheapest → most aggressive): conservative (start on pointerdown), moderate (start after ~200ms hover), eager (start as soon as the link is in the viewport), immediate (start on page load). Start with moderate — it captures most navigations without prerendering pages users never visit.

Caveats:

  • Bandwidth/CPU cost. Each prerender is roughly a full page load. Scope where carefully (href_matches patterns, exclude logout/checkout) and avoid immediate outside small sites.
  • Side effects fire early. Analytics, ads, and any code that runs on load will fire when the prerender starts, not when the user navigates. Gate side effects on the prerenderingchange event or document.prerendering.
  • Chromium-only. Safari and Firefox ignore the script — it's a progressive enhancement, never a regression.

LCP optimization checklist

markdown
- [ ] TTFB < 800ms (use CDN, edge caching)
- [ ] LCP image preloaded with fetchpriority="high"
- [ ] LCP image optimized (WebP/AVIF, correct size)
- [ ] Critical CSS inlined (< 14KB)
- [ ] No render-blocking JavaScript in <head>
- [ ] Fonts don't block text rendering (font-display: swap)
- [ ] LCP element in initial HTML (not JS-rendered)
- [ ] Speculation Rules added for likely-next navigations (moderate eagerness)

LCP element identification

javascript
// Find your LCP element
new PerformanceObserver((list) => {
  const entries = list.getEntries();
  const lastEntry = entries[entries.length - 1];
  console.log('LCP element:', lastEntry.element);
  console.log('LCP time:', lastEntry.startTime);
}).observe({ type: 'largest-contentful-paint', buffered: true });

INP: Interaction to Next Paint

INP measures responsiveness across ALL interactions (clicks, taps, key presses) during a page visit. It reports the worst interaction (at 98th percentile for high-traffic pages).

INP breakdown

Total INP = Input Delay + Processing Time + Presentation Delay

PhaseTargetOptimization
Input Delay< 50msReduce main thread blocking
Processing< 100msOptimize event handlers
Presentation< 50msMinimize rendering work

Common INP issues

1. Long tasks blocking main thread

javascript
// ❌ Long synchronous task
function processLargeArray(items) {
  items.forEach(item => expensiveOperation(item));
}

// ✅ Break into chunks and yield to the scheduler. scheduler.yield() is the
//    recommended modern API — its continuation is queued at a boosted
//    priority so the rest of your work resumes ahead of unrelated tasks,
//    while still letting the browser handle pending input first.
async function processLargeArray(items) {
  const CHUNK_SIZE = 100;
  for (let i = 0; i < items.length; i += CHUNK_SIZE) {
    items.slice(i, i + CHUNK_SIZE).forEach(expensiveOperation);

    if ('scheduler' in window && 'yield' in scheduler) {
      await scheduler.yield();
    } else {
      // Fallback for browsers without scheduler.yield (Safari, older Firefox).
      // setTimeout(0) yields but loses priority — your continuation may run
      // after unrelated tasks the browser picked up in between.
      await new Promise(r => setTimeout(r, 0));
    }
  }
}

2. Heavy event handlers

javascript
// ❌ All work in handler
button.addEventListener('click', () => {
  // Heavy computation
  const result = calculateComplexThing();
  // DOM updates
  updateUI(result);
  // Analytics
  trackEvent('click');
});

// ✅ Prioritize visual feedback, then yield before doing the heavy work
button.addEventListener('click', async () => {
  // 1. Immediate visual feedback (cheap DOM update)
  button.classList.add('loading');

  // 2. Yield so the browser can paint the loading state before we block
  if ('scheduler' in window && 'yield' in scheduler) {
    await scheduler.yield();
  }

  // 3. Now do the heavy work — the user already saw the click register
  const result = calculateComplexThing();
  updateUI(result);

  // 4. Lowest-priority work last, when the main thread is idle
  if ('requestIdleCallback' in window) {
    requestIdleCallback(() => trackEvent('click'));
  } else {
    setTimeout(() => trackEvent('click'), 0);
  }
});

3. Third-party scripts

javascript
// ❌ Eagerly loaded, blocks interactions
<script src="https://heavy-widget.com/widget.js"></script>

// ✅ Lazy loaded on interaction or visibility
const loadWidget = () => {
  import('https://heavy-widget.com/widget.js')
    .then(widget => widget.init());
};
button.addEventListener('click', loadWidget, { once: true });

4. Excessive re-renders (React/Vue)

javascript
// ❌ Re-renders entire tree
function App() {
  const [count, setCount] = useState(0);
  return (
    <div>
      <Counter count={count} />
      <ExpensiveComponent /> {/* Re-renders on every count change */}
    </div>
  );
}

// ✅ Memoized expensive components
const MemoizedExpensive = React.memo(ExpensiveComponent);

function App() {
  const [count, setCount] = useState(0);
  return (
    <div>
      <Counter count={count} />
      <MemoizedExpensive />
    </div>
  );
}

INP optimization checklist

markdown
- [ ] No tasks > 50ms on main thread
- [ ] Event handlers complete quickly (< 100ms)
- [ ] Visual feedback provided immediately
- [ ] Heavy work deferred with requestIdleCallback
- [ ] Third-party scripts don't block interactions
- [ ] Debounced input handlers where appropriate
- [ ] Web Workers for CPU-intensive operations

INP debugging

javascript
// Identify slow interactions. durationThreshold: 40 matches what the
// web-vitals library uses — 16 (one frame) fires on nearly every interaction
// and drowns the console; 40 surfaces interactions that are starting to feel
// sluggish without spamming.
new PerformanceObserver((list) => {
  for (const entry of list.getEntries()) {
    if (entry.duration > 200) {
      console.warn('Slow interaction:', {
        type: entry.name,
        duration: entry.duration,
        processingStart: entry.processingStart,
        processingEnd: entry.processingEnd,
        target: entry.target
      });
    }
  }
}).observe({ type: 'event', buffered: true, durationThreshold: 40 });

For field debugging across real users, prefer the web-vitals/attribution build of the web-vitals library — onINP() from that build attaches a LoAF (Long Animation Frame) breakdown identifying the longest script and the input/processing/presentation phase that ate the budget.


CLS: Cumulative Layout Shift

CLS measures unexpected layout shifts. A shift occurs when a visible element changes position between frames without user interaction.

CLS Formula: impact fraction × distance fraction

Common CLS causes

1. Images without dimensions

html
<!-- ❌ Causes layout shift when loaded -->
<img src="photo.jpg" alt="Photo">

<!-- ✅ Space reserved -->
<img src="photo.jpg" alt="Photo" width="800" height="600">

<!-- ✅ Or use aspect-ratio -->
<img src="photo.jpg" alt="Photo" style="aspect-ratio: 4/3; width: 100%;">

2. Ads, embeds, and iframes

html
<!-- ❌ Unknown size until loaded -->
<iframe src="https://ad-network.com/ad"></iframe>

<!-- ✅ Reserve space with min-height -->
<div style="min-height: 250px;">
  <iframe src="https://ad-network.com/ad" height="250"></iframe>
</div>

<!-- ✅ Or use aspect-ratio container -->
<div style="aspect-ratio: 16/9;">
  <iframe src="https://youtube.com/embed/..." 
          style="width: 100%; height: 100%;"></iframe>
</div>

3. Dynamically injected content

javascript
// ❌ Inserts content above viewport
notifications.prepend(newNotification);

// ✅ Insert below viewport or use transform
const insertBelow = viewport.bottom < newNotification.top;
if (insertBelow) {
  notifications.prepend(newNotification);
} else {
  // Animate in without shifting
  newNotification.style.transform = 'translateY(-100%)';
  notifications.prepend(newNotification);
  requestAnimationFrame(() => {
    newNotification.style.transform = '';
  });
}

4. Web fonts causing FOUT

css
/* ❌ Font swap shifts text */
@font-face {
  font-family: 'Custom';
  src: url('custom.woff2') format('woff2');
}

/* ✅ Optional font (no shift if slow) */
@font-face {
  font-family: 'Custom';
  src: url('custom.woff2') format('woff2');
  font-display: optional;
}

/* ✅ Or match fallback metrics */
@font-face {
  font-family: 'Custom';
  src: url('custom.woff2') format('woff2');
  font-display: swap;
  size-adjust: 105%; /* Match fallback size */
  ascent-override: 95%;
  descent-override: 20%;
}

5. Animations triggering layout

css
/* ❌ Animates layout properties */
.animate {
  transition: height 0.3s, width 0.3s;
}

/* ✅ Use transform instead */
.animate {
  transition: transform 0.3s;
}
.animate.expanded {
  transform: scale(1.2);
}

CLS optimization checklist

markdown
- [ ] All images have width/height or aspect-ratio
- [ ] All videos/embeds have reserved space
- [ ] Ads have min-height containers
- [ ] Fonts use font-display: optional or matched metrics
- [ ] Dynamic content inserted below viewport
- [ ] Animations use transform/opacity only
- [ ] No content injected above existing content

CLS debugging

javascript
// Track layout shifts
new PerformanceObserver((list) => {
  for (const entry of list.getEntries()) {
    if (!entry.hadRecentInput) {
      console.log('Layout shift:', entry.value);
      entry.sources?.forEach(source => {
        console.log('  Shifted element:', source.node);
        console.log('  Previous rect:', source.previousRect);
        console.log('  Current rect:', source.currentRect);
      });
    }
  }
}).observe({ type: 'layout-shift', buffered: true });

Measurement tools

Lab testing

  • Chrome DevTools → Performance panel, Lighthouse
  • WebPageTest → Detailed waterfall, filmstrip
  • Lighthouse CLI → npx lighthouse <url>

Field data (real users)

  • Chrome User Experience Report (CrUX) → BigQuery or API
  • Search Console → Core Web Vitals report
  • web-vitals library → Send to your analytics
javascript
import {onLCP, onINP, onCLS} from 'web-vitals';

function sendToAnalytics({name, value, rating}) {
  gtag('event', name, {
    event_category: 'Web Vitals',
    value: Math.round(name === 'CLS' ? value * 1000 : value),
    event_label: rating
  });
}

onLCP(sendToAnalytics);
onINP(sendToAnalytics);
onCLS(sendToAnalytics);

Framework quick fixes

Next.js

jsx
// LCP: Use next/image with priority
import Image from 'next/image';
<Image src="/hero.jpg" priority fill alt="Hero" />

// INP: Use dynamic imports
const HeavyComponent = dynamic(() => import('./Heavy'), { ssr: false });

// CLS: Image component handles dimensions automatically

React

jsx
// LCP: Preload in head
<link rel="preload" href="/hero.jpg" as="image" fetchpriority="high" />

// INP: Memoize and useTransition
const [isPending, startTransition] = useTransition();
startTransition(() => setExpensiveState(newValue));

// CLS: Always specify dimensions in img tags

Vue/Nuxt

vue
<!-- LCP: Use nuxt/image with preload -->
<NuxtImg src="/hero.jpg" preload loading="eager" />

<!-- INP: Use async components -->
<component :is="() => import('./Heavy.vue')" />

<!-- CLS: Use aspect-ratio CSS -->
<img :style="{ aspectRatio: '16/9' }" />

References

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Core Web Vitals AI skill do?

Optimize Core Web Vitals (LCP, INP, CLS) for better page experience and search ranking. Use when asked to "improve Core Web Vitals", "fix LCP", "reduce CLS", "optimize INP", "page experience optimization", or "fix layout shifts".

Why use Core Web Vitals on TypingMind?

Because you install it once and use it with any model. Core Web Vitals is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Core Web Vitals in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/NVlabs/Skill2Env/tree/main/SkillHub/skills/addy-osmani/core-web-vitals. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Core Web Vitals?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Core Web Vitals?

As many as you like. As long as a model supports skills, you can use Core Web Vitals with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Core Web Vitals AI skill free?

Yes. It is published on GitHub by NVlabs under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇