core-web-vitals

LCP fixes target four areas: server response time under 800ms, making the LCP resource discoverable in initial HTML with fetchpriority="high", inlining critical CSS, and avoiding client-side rendering of the LCP element. Measure with a trace first to identify which subpart dominates.

INP improvement starts by identifying which phase dominates: input delay, processing time, or presentation delay. Fixes include yielding long tasks with scheduler.yield(), prioritizing visible feedback, deferring third-party scripts, and moving CPU-heavy work to Web Workers.

Google measures at the 75th percentile of page visits. Good thresholds are LCP at or under 2.5 seconds, INP at or under 200 milliseconds, and CLS at or under 0.1. Values between good and poor thresholds are classified as needing improvement.

Yes, the Skill includes framework-specific quick fixes for Next.js, React, Vue/Nuxt, and Astro. Examples include using next/image with priority for LCP, useTransition and dynamic imports for INP, and explicit image dimensions for CLS.

Lab measurements capture a single controlled session while CrUX reports the p75 distribution of real user visits. A single lab value cannot be compared directly to a field p75, and field improvement only appears after new user visits accumulate.

Use prerendering for sites with predictable same-origin navigation journeys, starting with moderate eagerness. It is Chromium-only, each prerender costs roughly a full page load, and side effects like analytics fire early, so measure hit rate and server load before expanding.