Chrome for Developers has released Modern Web Guidance 1.0, a set of skills that plug into AI coding agents and steer them toward current web platform features instead of older JavaScript workarounds. In a Chrome for Developers blog post, engineer Philip Walton says the project left its early preview from Google I/O in May and is now a stable release. The official Chrome account says version 1.0 includes more than 200 expert guides, with Baseline browser-support data built in.
The release is a joint effort. Chrome’s announcement names Firefox, Microsoft Edge and MDN Web Docs as partners. Walton’s post says Firefox and Edge are joining Chrome “to support, improve cross-browser compatibility, and contribute to the project,” building on their shared work on Baseline.
What 1.0 adds
The skills sit in the agent you already use. Official docs list install paths for Google Antigravity, Cursor, GitHub Copilot, Claude Code, OpenAI’s Codex, JetBrains WebStorm, Gemini CLI, Grok, Kimi Code and others. A later Chrome post says you can add the package from a project directory with one command:
npx modern-web-guidance@latest installThe documentation says the CLI installer is the recommended route because it can apply automatic updates. Installing through some agent marketplaces, it notes, does not. Projects that already had a preview build can move to 1.0 with npx modern-web-guidance@latest update. The code is on GitHub under an Apache 2.0 licence, in the GoogleChrome/modern-web-guidance repository, which published a v1.0.0 release to match the announcement.
New in this version, Walton writes, is a modernisation workflow that scans existing HTML, CSS, JavaScript and package.json files for upgrades the team considers safe, then walks the agent through the changes. In this first release the scan uses static analysis on a limited set of patterns, with plans to widen the checks later. The post also describes a larger guide database, deeper review (including an accessibility review of user-facing guides, tested with real assistive technologies), and a public eval dashboard.
Chrome’s get-started guide says early results show, on average, a 37 percentage point improvement in adherence to modern best practices when agents use Modern Web Guidance. Those figures are Chrome’s own. The same page says results can vary by project, model, prompts and tools, and that the evals compare a guided run with an unguided control on the same task. The dashboard notes that task prompts don’t mention the skills, so the agent has to recognise when a guide applies.
How the guidance is meant to work
The docs describe a familiar failure mode: models often reach for custom JavaScript to solve problems that newer HTML or CSS can handle in less code, and they can be out of date on browser support. Guides cover use cases such as dialogs and popovers, CSS anchor positioning, container queries, Interaction to Next Paint, passkeys, and focus management. If a project doesn’t set a Baseline target in a file such as AGENTS.md or CLAUDE.md, the guidance defaults to Baseline Widely available. Features that are only newly or limited in support come with fallback instructions; guidance for volatile, single-engine features and origin-trial APIs is off unless you turn it on.
Sarah Drasner, Chrome’s area tech lead for AI and the web ecosystem, said coding models “get stuck in old training data,” and that the skills are vetted by industry experts for performance, accessibility, security and current platform features. MDN Web Docs replied to Chrome’s announcement to say it was glad to help shape the project “and make sure it works across every browser.”
What Chrome says teams are already doing with it
The launch post includes case studies from teams that used preview builds alongside Chrome DevTools for agents. Those claims come from Chrome and the named companies, and they haven’t been independently repeated here. Among them: MetaMask says it ran the skills as a review bot on pull requests to its browser extension, and that engineers fixed 43 findings in the first five weeks, mostly accessibility, HTML semantics and forms issues that linters missed. Japan’s note inc. says a 180-run evaluation on its production Next.js codebase raised pass rates on modern web checks from 61% to 94%. WeTV, Tencent’s streaming service, says an agent in its CI/CD pipeline now diagnoses and resolves 90% of failed performance checks.
Walton recommends pairing the skills with Chrome DevTools for agents, so the model can write against current guidance and then verify the result at runtime, including performance traces, reduced-motion checks and memory use. This is a separate release from this week’s Chrome 157 HTML-in-Canvas feature. Modern Web Guidance can also be installed in Google Antigravity, which recently gained a Knowledge Work layout.
Discover more from Sulat AI & Tech News
Subscribe to get the latest posts sent to your email.
