CiteStat/Wiki/Modern Web Guidance — Chrome's web platform guides for coding agents

Modern Web Guidance — Chrome's web platform guides for coding agents

Technical SEO

TL;DR

Modern Web Guidance is a set of guides ("skills") that gives AI coding agents current web platform practice and browser support data. Chrome released version 1.0 on 10 November 2026, in a post by Philip Walton, with Firefox and Edge joining the project. The catalog holds 208 guides in 17 categories, from performance and security to forms, accessibility and WebMCP. It ships with legacy-scan, a command-line tool that finds outdated patterns in your source code.

Explain Simply

Think of it as the house style manual a coding assistant reads before it touches your website: "on a sign-in form, mark the password field current-password", "do not hold the main thread for more than 50 ms". legacy-scan is the proofreader that circles the old habits already sitting in your code.

What ships

  • 208 guides in 17 categories: Visual Design, UI Atoms, UI Components, UI Behaviors, User Experience, Forms, HTML, CSS, JavaScript, WebAssembly, Performance, Accessibility, Security, Privacy, PWA, Built-in AI and WebMCP.
  • Each guide lives at developer.chrome.com/docs/modern-web-guidance/guides/<category>/<slug>, for example performance/identify-inp-causes.
  • legacy-scan reads HTML, CSS, JavaScript/TypeScript, framework components and package.json locally, with no LLM call.
npx modern-web-guidance@latest install
npx modern-web-guidance legacy-scan <path> --json

Version 1.0.0 of legacy-scan runs 35 deterministic rules over 33 guides, which Chrome's modernization workflow page puts at about 20% of the guides: only the ones where old code can be found without false positives. Chrome's modernization workflow page also says a rule ships only at 80% static matcher precision and 90% after the agent verifies the match, measured on unseen code from about 45,000 open-source repositories. There is no --fix flag: the agent retrieves the guide (npx modern-web-guidance retrieve <guide-id>), refactors, then scans and tests again.

Guides with an SEO or AI-search angle

  • Performance: break-up-long-tasks and identify-inp-causes (a main-thread task over 50 ms hurts INP), optimize-image-priority (the LCP image), optimize-script-priority, improve-next-page-load-performance (speculation rules), lazy-load-video-and-audio.
  • Security: security (CSP, HSTS, nosniff), secure-external-subresources (SRI, Referrer-Policy), trusted-types.
  • Forms: forms, autofill-sign-in-form, autofill-sign-up-form (autocomplete tokens).
  • Accessibility: accessibility, respect-os-text-scale (opt into OS text scaling with meta name=text-scale).
  • Visual design: dark-mode (the color-scheme meta tag).
  • PWA: install-web-app (the web app manifest).
  • WebMCP: webmcp, agentic-forms, agentic-javascript-tools.

Our own run: legacy-scan 1.0.0 on our web app found 2 issues, a sign-in password field without autocomplete="current-password" and a sign-up one without "new-password". We added the tokens; the re-scan found 0.

⚡Why It Matters

The guides are written for coding agents, not for search engines, so following them is not a ranking signal in itself. They matter because several guides fix what search does measure: INP and LCP are Core Web Vitals, HTTPS and security headers shape trust, a page that blocks zoom fails accessibility checks, and WebMCP decides whether a browser agent can finish a task on your site. More website code is now written by agents; this is the reference the Chrome, Firefox and Edge teams want those agents to follow.

🔍How CiteStat Checks

CiteStat scans the page a visitor receives (HTML, response headers, Lighthouse), not your source code, so it covers the guides that leave a mark in a served page. Checked today: the LCP image's priority (fetchpriority, no lazy loading on the hero), render-blocking scripts and preloads, long main-thread work and INP (Lighthouse Total Blocking Time, CrUX field INP), third-party scripts without an integrity (SRI) hash, security headers (CSP, HSTS, nosniff), a viewport that disables zoom (a WCAG 1.4.4 check, separate from the text-scale guide), image alt text and the page language, the web app manifest, and WebMCP tools. Partly: lazy loading is scored for images and only reported for video and audio; font-display is read, size-adjust fallbacks are not. Scored: autocomplete on sign-in and sign-up fields (a warning, weight 2, only on pages that have such a field). Reported, not scored: speculation rules, Trusted Types in the CSP header, the color-scheme meta tag, the text-scale meta tag, and video and audio preload. For your own source code, run legacy-scan.

CiteStat Score Impact

No rule links here: this page maps the guides, it is not a check. The checks it names are scored on their own wiki pages.

Fix Difficulty

Editor: Not relevant | Developer: Easy (install, scan) to Medium (refactor) | AI Agent: Easy