Agent skills and an offline search CLI that feed coding agents current web-platform guidance and Baseline browser-compatibility data.
- License
- CC BY 4.0 (content), Apache 2.0 (code samples)
- Try it
npx modern-web-guidance@latest install
Google's Chrome team released Modern Web Guidance 1.0 on November 10, taking its open-source skill pack for AI coding agents out of preview. The project feeds agents like Claude Code, Codex, Cursor and Copilot curated guidance on current browser features, so they reach for a native <dialog>, container queries or the Popover API instead of piling on old JavaScript workarounds. With 1.0, Microsoft Edge and Mozilla Firefox join Chrome in backing the project, and it gains a scanner that hunts for legacy patterns in an existing codebase.
Why it matters: coding agents learn from years of old code, so they often write the web as it was in 2019. If you ship front-end code an agent wrote, this is a free, one-command way to change what it produces, and it now has more than one browser vendor behind it.
What's in 1.0
According to the v1.0.0 release notes:
| Item | 1.0 |
|---|---|
| Use-case guides | 203 |
| Web platform features covered | 250 |
| New command | legacy-scan, a static check that maps legacy patterns to modern replacements |
| Search engine | Zero-dependency embedding model plus BM25, replacing TensorFlow.js |
| npm download size | 8.8 MB, down from 23.5 MB |
| Guides updated | 98, mostly for accessibility |
Search runs offline, with no API keys or network calls, according to the project README. Guides for features that only one browser engine supports can still be found in search. To retrieve them, you have to opt in with a .mwgrc file. By default the guidance targets Baseline "widely available" features. LogRocket notes that you can set a different target, such as Baseline 2024, in files like AGENTS.md or CLAUDE.md.
The Chrome blog post, written by Philip Walton, says each guide is signed off by a tech lead for its content area, and guides for user-facing features also get an accessibility review. The supported tools are Claude Code, Codex, Cursor, Antigravity, Copilot, WebStorm, Kimi Code and Grok Build, plus the standalone CLI. Guide content is licensed CC BY 4.0, and code samples are Apache 2.0.
Governance: who's at the table
Search Engine Journal reports that a governance document, posted November 9, creates six owner seats: two for Chrome, one each for Edge and Mozilla, and two for independent experts, one of which is still open. Apple has no seat. For a project that tells agents what is safe to use across browsers, that gap is worth noting.
Do the numbers hold up?
Google publishes a nightly eval dashboard that runs agents from Anthropic, OpenAI and Google with and without the guidance. The sources don't all report the same figures:
- Search Engine Journal, reading the dashboard runs from November 8 (199–215 tasks per agent): guided agents passed 88% of automated checks, against 56% unguided.
- The release notes report gains of 35 to 38 percentage points across three agent and model combinations.
- The Show HN author (posting as paulirish) cited about 1,700 Playwright and AST-based assertions across 200 tasks, with pass rates rising from 55% to 92%. He also said
legacy-scanaverages 92% precision on held-out repos.
The Chrome blog also lists customer results. Japan's note inc. says pass rates on its modern-web checks rose from 61% to 94% over 180 runs. MetaMask runs the guidance in a pull-request review bot. Search Engine Journal points out that MetaMask's results were mixed: of 102 issues flagged, 43 were fully resolved and 16 were declined or false positives, while 20 went unaddressed and 20 had unknown outcomes.
The caveats are real. Google runs the dashboard. Its accessibility checks are limited, and the dashboard itself says they are not a full WCAG audit. Search Engine Journal also found that guided runs were usually slower and used more tokens. Accessibility consultant Adrian Roselli criticised the May preview over output that did not meet WCAG, though he tested the preview, not 1.0. And the scanner covers only a limited set of patterns for now.
Try it
Node.js 18 or later is required. To install for your agent:
npx modern-web-guidance@latest install
Or try it without installing:
npx modern-web-guidance@latest search "accessible modal dialog"
npx modern-web-guidance@latest legacy-scan .
If you already used the preview, run npx modern-web-guidance@latest update.
Who should care
This is most useful for front-end teams that let agents write UI code, or review a lot of agent-written code. The legacy-scan command is a cheap first pass on an older codebase. Treat the accessibility output as a starting point, not a sign-off, and watch your token bills: extra context isn't free. Also watch whether the open independent seat gets filled and whether Apple ever joins. A multi-vendor guide is more trustworthy than a single-vendor one.
Sources
- Chrome for Developers — Modern Web Guidance 1.0: The web platform expertise your coding agent is missing developer.chrome.com
- GitHub — GoogleChrome/modern-web-guidance release v1.0.0 github.com
- Search Engine Journal — Google Releases Modern Web Guidance 1.0 For AI Coding Agents searchenginejournal.com
- Hacker News — Show HN: Get your coding agent off ~2019 JavaScript, onto modern web primitives news.ycombinator.com
- LogRocket — Chrome's Modern Web Guidance for AI coding agents blog.logrocket.com