---
title: "Core Web Vitals: review, setup & limitations · Undominated.ai"
canonical: https://undominated.ai/skills/addyosmani-core-web-vitals/
description: "Diagnoses LCP, INP and CLS from field evidence first — CrUX p75 with a labelled origin fallback, then a browser trace — and routes to per-metric references, with explicit rules against claiming field improvement from a lab run."
---

# Core Web Vitals: review, setup & limitations · Undominated.ai

> Diagnoses LCP, INP and CLS from field evidence first — CrUX p75 with a labelled origin fallback, then a browser trace — and routes to per-metric references, with explicit rules against claiming field improvement from a lab run.

[← Explore all skills](/skills/)

WEB PERFORMANCE / Addy Osmani

# Core Web Vitals

Diagnoses LCP, INP and CLS from field evidence first — CrUX p75 with a labelled origin fallback, then a browser trace — and routes to per-metric references, with explicit rules against claiming field improvement from a lab run.

 [See setup guidance ↓](#setup)[Original source ↗](https://github.com/addyosmani/web-quality-skills/tree/afa8da942115f2961fdbfa80807ea0b232ff6c00/skills/core-web-vitals)

SOURCE REVIEW

 Reviewed 2026-10-15
 Evidence 7 linked sources
 Publisher Addy Osmani
 Licence [MIT ↗](https://github.com/addyosmani/web-quality-skills/blob/afa8da942115f2961fdbfa80807ea0b232ff6c00/LICENSE)
 Revision afa8da942115
 [Read what was—and wasn’t—checked ↓](#review)

“Optimize Core Web Vitals (LCP, INP, CLS) for better page experience using field and lab evidence. …”

 [Addy Osmani · upstream description ↗](https://github.com/addyosmani/web-quality-skills/blob/afa8da942115f2961fdbfa80807ea0b232ff6c00/skills/core-web-vitals/SKILL.md) Our analysis follows below.

01 / THE REASONING

## Why this made the selection.

 - Measurement discipline is the spine: CrUX p75 first with a labelled fallback when page-level data is missing, trace analysis of only the failing metric's insights, and re-measurement after the fix with field verification explicitly deferred until new user data exists.
- The Speculation Rules section is the careful version — eagerness triggers per Chrome version, bandwidth cost, side effects firing at prerender time and gated on document.prerendering — instead of a copy-paste snippet.
- Per-metric references (LCP.md, INP.md, CLS.md) exist in the skill directory, and the skill forbids comparing a single lab value with a field p75 as if they were equivalent samples.

### A good fit for

 - Fixing a failing LCP, INP or CLS with trace-based diagnosis rather than generic advice
- Deciding whether prerendering likely-next pages is safe for a journey
- Setting up first-party RUM attribution for the three vitals

### Weigh up before choosing

 - Cross-links the sibling performance skill's references (../performance/references/MEASUREMENT.md and RUM.md); installing core-web-vitals alone leaves those two links dangling.
- Chrome-centric: CrUX, DevTools traces and Speculation Rules are Chromium surfaces; Safari and Firefox get progressive-enhancement notes only.
- Framework quick fixes for Next.js, React and Vue/Nuxt are one-liners — starting points, not measured guidance.
- A changed lab measurement does not establish immediate improvement in CrUX or first-party RUM; those field datasets require new user visits.

02 / THE REVIEW RECORD

## What we actually inspected.

Source review has boundaries. A clear record is more useful than a “safe” badge.

### Material inspected

 - skills/core-web-vitals/SKILL.md (complete, 228 lines)
- skills/core-web-vitals/references directory (LCP.md, INP.md, CLS.md)
- cross-link targets in the sibling performance skill
- repository LICENSE (MIT)
- README.md (installation command audit at the existing pin)

### Our findings

 - The metric table states the good, needs-work and poor thresholds and that Google measures at the 75th percentile, and the workflow refuses to claim a metric is failing without runtime evidence.
- The LCP section prefers a discoverable img with fetchpriority over speculative preloads and says preload only when the trace shows late discovery — the ordering most guides get wrong.
- The measurement-sources table separates trace, CrUX, Lighthouse and RUM uses and forbids routing performance through the DevTools lighthouse_audit capability.

### Not established by this review

 - No trace was recorded and no metric was measured; the workflow was read, not run.
- Speculation Rules behaviour claims were checked against the linked Chrome documentation, not reproduced in a browser.
- Not executed: it was not installed or run.

The review applies to the material and revision named here. A newer upstream release can change its behavior.

03 / PUT IT TO WORK

## Add a skill to your workflow.

[Upstream setup instructions ↗](https://github.com/vercel-labs/skills/blob/7407f3893ad4dceab546ac002c3ef806e4000c73/README.md)

DOCUMENTED COMMAND

 npx skills add addyosmani/web-quality-skills/skills/core-web-vitals --skill core-web-vitals Copy command ↗

Run in your terminal to choose your AI tools, project or global scope, and installation method. This short command follows the current upstream version; the review covers the revision shown above.

 Optional: install the reviewed revision

Keeps the original source and installer version where recorded, while leaving your AI tools and installation scope for you to choose.

 npx skills add https://github.com/addyosmani/web-quality-skills/tree/afa8da942115f2961fdbfa80807ea0b232ff6c00/skills/core-web-vitals --skill core-web-vitals Copy reviewed-revision command ↗ - Install the complete skill directory; the per-metric references must travel with the SKILL.md.
- For the full measurement workflow, install the sibling performance skill too — two cross-links point into its references.

### Before you start

 - Node.js/npm for the installer
- An agent that supports the Agent Skills format
- A runnable page or CrUX data; a browser tool that records performance traces when available

### Compatibility

Agent Skills-compatible coding agents · Any web page that can be measured in a browser or via CrUX · Chrome DevTools MCP referenced for traces; Lighthouse CLI as lab fallback

### Reads front-end code and measurement data; writes performance fixes

 - Read page source, resource loading and field or lab measurement data
- Write HTML, CSS, JavaScript and image-loading changes
- Run measurement tooling when the user asks

### Cost model

MIT-licensed instruction package; measurement tooling and browser time are the only costs.

04 / FOLLOW THE EVIDENCE

## The source trail.

Our notes are separate from the original resource. Check upstream before adopting a new version.

 - [Skill definition at reviewed revision ↗](https://github.com/addyosmani/web-quality-skills/blob/afa8da942115f2961fdbfa80807ea0b232ff6c00/skills/core-web-vitals/SKILL.md) Checked 2026-10-07 https://github.com/addyosmani/web-quality-skills/blob/afa8da942115f2961fdbfa80807ea0b232ff6c00/skills/core-web-vitals/SKILL.md Supports: summary, upstreamDescription, whySelected, bestFor, limitations, review, access
- [Per-metric reference directory (LCP.md, INP.md, CLS.md) ↗](https://github.com/addyosmani/web-quality-skills/tree/afa8da942115f2961fdbfa80807ea0b232ff6c00/skills/core-web-vitals/references) Checked 2026-10-07 https://github.com/addyosmani/web-quality-skills/tree/afa8da942115f2961fdbfa80807ea0b232ff6c00/skills/core-web-vitals/references Supports: review, whySelected
- [Sibling performance skill (cross-link target for MEASUREMENT.md and RUM.md) ↗](https://github.com/addyosmani/web-quality-skills/tree/afa8da942115f2961fdbfa80807ea0b232ff6c00/skills/performance) Checked 2026-10-07 https://github.com/addyosmani/web-quality-skills/tree/afa8da942115f2961fdbfa80807ea0b232ff6c00/skills/performance Supports: limitations, install
- [Repository licence (MIT) ↗](https://github.com/addyosmani/web-quality-skills/blob/afa8da942115f2961fdbfa80807ea0b232ff6c00/LICENSE) Checked 2026-10-07 https://github.com/addyosmani/web-quality-skills/blob/afa8da942115f2961fdbfa80807ea0b232ff6c00/LICENSE Supports: license
- [Skills installer source formats and selective installation ↗](https://github.com/vercel-labs/skills/blob/7407f3893ad4dceab546ac002c3ef806e4000c73/README.md) Checked 2026-10-07 https://github.com/vercel-labs/skills/blob/7407f3893ad4dceab546ac002c3ef806e4000c73/README.md Supports: install
- [Definition excerpt rechecked at the existing revision ↗](https://raw.githubusercontent.com/addyosmani/web-quality-skills/afa8da942115f2961fdbfa80807ea0b232ff6c00/skills/core-web-vitals/SKILL.md) Checked 2026-10-15 https://raw.githubusercontent.com/addyosmani/web-quality-skills/afa8da942115f2961fdbfa80807ea0b232ff6c00/skills/core-web-vitals/SKILL.md Supports: limitations, upstreamDescription, review
- [Installation-source command audit at the existing revision ↗](https://raw.githubusercontent.com/vercel-labs/skills/7407f3893ad4dceab546ac002c3ef806e4000c73/README.md) Checked 2026-10-15 https://raw.githubusercontent.com/vercel-labs/skills/7407f3893ad4dceab546ac002c3ef806e4000c73/README.md Supports: review

KEEP COMPARING

## Other approaches to consider.

Related by category or shared topics. These are alternatives to inspect, not a measured quality order.

 [### Cloudflare Web Performance Audit ↗ Builds a performance review from browser traces, request details and relevant source instead of generic optimization advice.](/skills/cloudflare-web-perf/)[### CPU Profile Analysis ↗ Walks an agent through analyzing V8 CPU profiles (.cpuprofile) and Chrome DevTools trace files: parsing large files safely, reconstructing call trees from sampling data, measuring self versus total time, and comparing code paths.](/skills/microsoft-cpu-profile-analysis/)[### Heap Snapshot Analysis ↗ Investigates memory leaks from V8 heap snapshots: parsing multi-hundred-megabyte .heapsnapshot files, comparing before/after object deltas, and tracing retainer paths while skipping the weak edges and debugger artifacts that produce false suspects.](/skills/microsoft-heap-snapshot-analysis/)

 [AI Tools ↗](/tools/)[Skills ↗](/skills/)[Agents ↗](/agents/)[MCP Servers ↗](/mcp-servers/)[Workflows ↗](/workflows/)

## Continue your investigation

 - [Choose a toolkit](/tools/)
- [Inspect execution hosts](/agents/)
- [Check tool access](/mcp-servers/)
- [Check project dependencies](/check-repo/)
