---
title: Web Perf
category: product
entity_type: skill
price: $15
canonical: https://forgehouse.ai/skills/web-perf/
lang: en
hreflang_alt: https://forgehouse.ai/tr/skiller/web-perf/
last_updated: 2026-06-20
---

# Web Perf

> Analyzes web performance using Chrome DevTools MCP.

A Chrome DevTools-driven web performance audit that measures Core Web Vitals (FCP, LCP, TBT, CLS, Speed Index) and pinpoints render-blocking resources, network dependency chains, layout-shift culprits and caching gaps. It quantifies impact in real milliseconds and prioritizes the few high-leverage fixes that move the needle. You get specific, actionable recommendations like 'compress the 450KB hero to WebP' instead of vague 'optimize images'.

## Use cases
- Auditing or profiling page load performance
- Diagnosing a poor Lighthouse or PageSpeed score
- Finding the true LCP element and its breakdown
- Tracing network waterfalls for serial dependency chains
- Identifying invisible CLS sources (font swap, ad slots)
- Analyzing the codebase for tree-shaking and dead code

## Benefits
- Focus effort on the 3-5 fixes that deliver 80% of the speed gain (Pareto)
- Quantify every recommendation in estimated milliseconds saved
- Separate lab data from real-user field data to avoid false confidence
- Skip zero-impact non-issues and recognize when a site is already excellent

## What’s included
- 5-phase audit (trace → CWV → network → a11y → codebase)
- Core Web Vitals thresholds (good/needs-improvement/poor) for each metric
- Render-blocking, preload and caching analysis with DevTools insights
- Framework/bundler detection (Webpack, Vite, Next.js, Astro and more)
- Top-5 field tricks for LCP misdiagnosis, invisible CLS and third-party TBT
- Prioritized output with Core Web Vitals summary and code-snippet fixes

## Who it’s for
Performance engineers and front-end developers who need a precise, impact-quantified path to faster page loads and better Core Web Vitals.

## How it runs
Every finding ships with an estimated gain in milliseconds, and only the top three to five make the final list. The five-phase trace through Chrome DevTools earns those numbers:
1. Verifies the chrome-devtools MCP is reachable first (navigate_page or performance_start_trace), then loads the target page and records a cold-load trace with reload enabled.
2. Extracts Core Web Vitals from the trace via performance_analyze_insight: LCPBreakdown (TTFB vs resource load vs render delay), CLSCulprits (undimensioned images, injected content, font swaps), RenderBlocking and DocumentLatency, each scored against the good/needs-improvement/poor thresholds (LCP 2.5s, INP 200ms, CLS 0.1, TTFB 800ms).
3. Lists every network request and reads the waterfall: render-blocking head resources, serial dependency chains, missing preloads, weak Cache-Control headers, oversized bundles, and disputed preconnects verified by counting actual requests to that origin.
4. Takes a verbose accessibility snapshot and flags duplicate ARIA IDs, WCAG AA contrast failures, missing focus indicators and unnamed interactive elements.
5. If codebase access exists, detects the framework and bundler from config files, then checks tree-shaking, barrel files, wholesale lodash/moment imports, oversized core-js polyfills and minification settings.
6. Outputs only the top 3 to 5 highest-impact issues, each with a concrete fix and an estimated gain in milliseconds; zero-impact findings are noted but never recommended as work.

## FAQ
### Do I need Chrome DevTools MCP connected for this to work?
Yes. The audit drives Chrome DevTools directly to record traces, measure FCP, LCP, TBT, CLS and Speed Index, and walk network waterfalls. Without that connection there is nothing to measure; it is not a static code-only analyzer.

### How does it decide which fixes to recommend first?
Every finding is quantified in estimated milliseconds saved, then Pareto-prioritized down to the 3-5 fixes that deliver most of the speed gain. That is why you get 'compress the 450KB hero to WebP' instead of a vague 'optimize images' bullet list.

### Does it report real-user field data from my actual visitors?
No. Measurements come from lab traces in DevTools runs; the audit deliberately separates lab from field data to avoid false confidence, but collecting real-user monitoring from your traffic is a separate setup it does not do.

## Price
$15, one-time, no subscription. VAT included.

Related guide: [AI code review and developer workflow](https://forgehouse.ai/guides/ai-code-review/)