Web performance audit
Audits a web page's loading performance through Chrome DevTools MCP: a trace, Core Web Vitals against thresholds, network, an accessibility snapshot and bundle config, with ranked fixes. Use when a page is slow or its Lighthouse score is low.
web-perf- Author
- Cloudflare
- Source
- cloudflare/skills/skills/web-perf @ 5108325
- License
- Apache-2.0
About this skill
What it does
The skill runs a performance audit of a web page through the Chrome DevTools MCP server and reports Core Web Vitals with concrete fixes. Without trace tools it falls back to source and network analysis and says which measurements are missing. SKILL.md gives the server entry (npx -y chrome-devtools-mcp@latest); the agent edits MCP config only within the scope you allowed.
The audit has five phases, tracked as a checklist. A trace with reload captures a cold load. performance_analyze_insight breaks it down by insight — LCPBreakdown, CLSCulprits, RenderBlocking, DocumentLatency, NetworkRequestsDepGraph — rated against thresholds for TTFB, FCP, LCP, INP, TBT, CLS and Speed Index. The network pass looks for render-blocking resources, request chains, missing preloads, weak caching headers, large payloads and preconnects to origins that got no requests. An accessibility snapshot flags only high-level gaps. Given the code, the agent finds the bundler or framework from config files and checks tree-shaking, barrel files, wholesale lodash or moment imports, polyfills, minification, compression and production source maps.
The guidelines keep the report honest: verify before recommending removal, skip issues with 0 ms estimated impact, name the file and its size, and say plainly when a site is already fast. The output is a vitals table, ranked issues with impact, recommendations with code or config, and codebase findings. references/skill-feedback.md handles reports about the skill itself.
When to use it, and when not
Use it for a slow page, a low Lighthouse score or Core Web Vitals that need explaining.
It is a lab audit of one page load: no real-user data, no backend profiling and no fix-and-remeasure loop.
How to run it
The install command below puts the folder in place. Add the Chrome DevTools MCP server to your agent; without it the skill has no traces. Then:
- Claude Code:
/web-perf, or ask for the skill by name. - Codex:
$web-perf. - Cursor: type
/in the Agent chat and pickweb-perf.
Example: /web-perf Audit http://localhost:5173/pricing and tell me what delays LCP; the code is in this repo.
How it differs from similar skills
Browser testing with Chrome DevTools MCP drives the same server for UI bugs, network issues and traces, with strict rules for treating page content as untrusted; this skill is a fixed audit with thresholds and a report format. Performance optimization measures, fixes one thing, re-measures and reverts what does not help. Vercel cost and performance tuning starts from production metrics of a Vercel project. Performance review reads code for N+1 queries and hot loops. React and Next.js performance rules checks React code against 70 rules ranked by impact and measures nothing.
Install
Run one command — it saves SKILL.md and the files it uses into your agent's skills folder. Claude Code and Codex pick up a new skill without a restart; if it does not show up, restart the agent.
curl -fsSL --create-dirs -o ~/.claude/skills/web-perf/SKILL.md https://anymodel.org/en/skills/cloudflare/web-perf/SKILL.md -o ~/.claude/skills/web-perf/LICENSE https://anymodel.org/en/skills/cloudflare/web-perf/LICENSE -o ~/.claude/skills/web-perf/references/skill-feedback.md https://anymodel.org/en/skills/cloudflare/web-perf/references/skill-feedback.md
curl -fsSL --create-dirs -o ~/.agents/skills/web-perf/SKILL.md https://anymodel.org/en/skills/cloudflare/web-perf/SKILL.md -o ~/.agents/skills/web-perf/LICENSE https://anymodel.org/en/skills/cloudflare/web-perf/LICENSE -o ~/.agents/skills/web-perf/references/skill-feedback.md https://anymodel.org/en/skills/cloudflare/web-perf/references/skill-feedback.md
foreach ($f in 'SKILL.md','LICENSE','references/skill-feedback.md') { $p = Join-Path "$HOME\.claude\skills\web-perf" $f; New-Item -ItemType Directory -Force (Split-Path $p) | Out-Null; Invoke-WebRequest "https://anymodel.org/en/skills/cloudflare/web-perf/$f" -UseBasicParsing -OutFile $p }foreach ($f in 'SKILL.md','LICENSE','references/skill-feedback.md') { $p = Join-Path "$HOME\.agents\skills\web-perf" $f; New-Item -ItemType Directory -Force (Split-Path $p) | Out-Null; Invoke-WebRequest "https://anymodel.org/en/skills/cloudflare/web-perf/$f" -UseBasicParsing -OutFile $p }Any other agent: save the file into its skills folder, or give it the link to the file: https://anymodel.org/en/skills/cloudflare/web-perf/SKILL.md
SKILL.md
Web Performance Audit
Your knowledge of web performance metrics, thresholds, and tooling APIs may be outdated. Prefer retrieval over pre-training when citing specific numbers or recommendations.
Retrieval Sources
| Source | How to retrieve | Use for |
|---|---|---|
| web.dev | https://web.dev/articles/vitals |
Core Web Vitals thresholds, definitions |
| Chrome DevTools docs | https://developer.chrome.com/docs/devtools/performance |
Tooling APIs, trace analysis |
| Lighthouse scoring | https://developer.chrome.com/docs/lighthouse/performance/performance-scoring |
Score weights, metric thresholds |
FIRST: Verify MCP Tools Available
Discover available browser and performance tools before starting. Use the capabilities available for the requested audit. If trace tools are unavailable, continue any useful source or network analysis and state which measurements could not be collected.
If the user wants Chrome DevTools MCP setup, consult its installation guide and use the latest package version. Only change MCP configuration when setup is within the user's authorized scope; otherwise ask first. For clients using command and args, an example server entry is:
"chrome-devtools": {
"command": "npx",
"args": ["-y", "chrome-devtools-mcp@latest"]
}
Key Guidelines
- Be assertive: Verify claims by checking network requests, DOM, or codebase—then state findings definitively.
- Verify before recommending: Confirm something is unused before suggesting removal.
- Quantify impact: Use estimated savings from insights. Don't prioritize changes with 0ms impact.
- Skip non-issues: If render-blocking resources have 0ms estimated impact, note but don't recommend action.
- Be specific: Say "compress hero.png (450KB) to WebP" not "optimize images".
- Prioritize ruthlessly: A site with 200ms LCP and 0 CLS is already excellent—say so.
Quick Reference
| Task | Tool Call |
|---|---|
| Load page | navigate_page(url: "...") |
| Start trace | performance_start_trace(autoStop: true, reload: true) |
| Analyze insight | performance_analyze_insight(insightSetId: "...", insightName: "...") |
| List requests | list_network_requests(resourceTypes: ["Script", "Stylesheet", ...]) |
| Request details | get_network_request(reqid: <id>) |
| A11y snapshot | take_snapshot(verbose: true) |
Workflow
Copy this checklist to track progress:
Audit Progress:
- [ ] Phase 1: Performance trace (navigate + record)
- [ ] Phase 2: Core Web Vitals analysis (includes CLS culprits)
- [ ] Phase 3: Network analysis
- [ ] Phase 4: Accessibility snapshot
- [ ] Phase 5: Codebase analysis (skip if third-party site)
Phase 1: Performance Trace
Navigate to the target URL:
navigate_page(url: "<target-url>")Start a performance trace with reload to capture cold-load metrics:
performance_start_trace(autoStop: true, reload: true)Wait for trace completion, then retrieve results.
Troubleshooting:
- If trace returns empty or fails, verify the page loaded correctly with
navigate_pagefirst - If insight names don't match, inspect the trace response to list available insights
Phase 2: Core Web Vitals Analysis
Use performance_analyze_insight to extract key metrics.
Note: Insight names may vary across Chrome DevTools versions. If an insight name doesn't work, check the insightSetId from the trace response to discover available insights.
Common insight names:
| Metric | Insight Name | What to Look For |
|---|---|---|
| LCP | LCPBreakdown |
Time to largest contentful paint; breakdown of TTFB, resource load, render delay |
| CLS | CLSCulprits |
Elements causing layout shifts (images without dimensions, injected content, font swaps) |
| Render Blocking | RenderBlocking |
CSS/JS blocking first paint |
| Document Latency | DocumentLatency |
Server response time issues |
| Network Dependencies | NetworkRequestsDepGraph |
Request chains delaying critical resources |
Example:
performance_analyze_insight(insightSetId: "<id-from-trace>", insightName: "LCPBreakdown")
Key thresholds (good/needs-improvement/poor):
- TTFB: < 800ms / < 1.8s / > 1.8s
- FCP: < 1.8s / < 3s / > 3s
- LCP: < 2.5s / < 4s / > 4s
- INP: < 200ms / < 500ms / > 500ms
- TBT: < 200ms / < 600ms / > 600ms
- CLS: < 0.1 / < 0.25 / > 0.25
- Speed Index: < 3.4s / < 5.8s / > 5.8s
Phase 3: Network Analysis
List all network requests to identify optimization opportunities:
list_network_requests(resourceTypes: ["Script", "Stylesheet", "Document", "Font", "Image"])
Look for:
- Render-blocking resources: JS/CSS in
<head>withoutasync/defer/mediaattributes - Network chains: Resources discovered late because they depend on other resources loading first (e.g., CSS imports, JS-loaded fonts)
- Missing preloads: Critical resources (fonts, hero images, key scripts) not preloaded
- Caching issues: Missing or weak
Cache-Control,ETag, orLast-Modifiedheaders - Large payloads: Uncompressed or oversized JS/CSS bundles
- Unused preconnects: If flagged, verify by checking if ANY requests went to that origin. If zero requests, it's definitively unused—recommend removal. If requests exist but loaded late, the preconnect may still be valuable.
For detailed request info:
get_network_request(reqid: <id>)
Phase 4: Accessibility Snapshot
Take an accessibility tree snapshot:
take_snapshot(verbose: true)
Flag high-level gaps:
- Missing or duplicate ARIA IDs
- Elements with poor contrast ratios (check against WCAG AA: 4.5:1 for normal text, 3:1 for large text)
- Focus traps or missing focus indicators
- Interactive elements without accessible names
Phase 5: Codebase Analysis
Skip if auditing a third-party site without codebase access.
Analyze the codebase to understand where improvements can be made.
Detect Framework & Bundler
Search for configuration files to identify the stack:
| Tool | Config Files |
|---|---|
| Webpack | webpack.config.js, webpack.*.js |
| Vite | vite.config.js, vite.config.ts |
| Rollup | rollup.config.js, rollup.config.mjs |
| esbuild | esbuild.config.js, build scripts with esbuild |
| Parcel | .parcelrc, package.json (parcel field) |
| Next.js | next.config.js, next.config.mjs |
| Nuxt | nuxt.config.js, nuxt.config.ts |
| SvelteKit | svelte.config.js |
| Astro | astro.config.mjs |
Also check package.json for framework dependencies and build scripts.
Tree-Shaking & Dead Code
- Webpack: Check for
mode: 'production',sideEffectsin package.json,usedExportsoptimization - Vite/Rollup: Tree-shaking enabled by default; check for
treeshakeoptions - Look for: Barrel files (
index.jsre-exports), large utility libraries imported wholesale (lodash, moment)
Unused JS/CSS
- Check for CSS-in-JS vs. static CSS extraction
- Look for PurgeCSS/UnCSS configuration (Tailwind's
contentconfig) - Identify dynamic imports vs. eager loading
Polyfills
- Check for
@babel/preset-envtargets anduseBuiltInssetting - Look for
core-jsimports (often oversized) - Check
browserslistconfig for overly broad targeting
Compression & Minification
- Check for
terser,esbuild, orswcminification - Look for gzip/brotli compression in build output or server config
- Check for source maps in production builds (should be external or disabled)
Output Format
Present findings as:
- Core Web Vitals Summary - Table with metric, value, and rating (good/needs-improvement/poor)
- Top Issues - Prioritized list of problems with estimated impact (high/medium/low)
- Recommendations - Specific, actionable fixes with code snippets or config changes
- Codebase Findings - Framework/bundler detected, optimization opportunities (omit if no codebase access)
Skill feedback
If the user reports that this skill gave incorrect or outdated guidance, or is missing information, read references/skill-feedback.md.
Connect through AnyModel API
This skill runs on whatever model your agent uses. Point Claude Code, Codex or Cursor at AnyModel to run it on Claude, GPT, Gemini and more with one key — pay per token, no subscription.
AnyModel