Chrome, Edge, Firefox and MDN teams launch Modern Web Guidance for coding agents

What launched

Google's Chrome organisation on GitHub has released Modern Web Guidance, a set of skills designed to keep AI coding agents current with the web platform. Announced through a Hacker News Show HN post on 10 November, the project is stewarded by the Google Chrome, Microsoft Edge, Mozilla Firefox and MDN teams alongside the web development community, according to its README.

Installing is a single command, npx modern-web-guidance@latest install, which runs an interactive wizard that sets the guidance up for an agent. The tool can also be sampled without installing anything: search finds guides relevant to a task, retrieve pulls a specific document by its ID, and legacy-scan walks an existing codebase and flags legacy patterns that could be modernised.

The problem it targets

The project's rationale is direct: coding agents reach for old solutions because their training data contains a huge amount of legacy code, so they produce needlessly heavy JavaScript for tasks the platform now handles natively. Even when a model knows a modern API exists, the README notes, it often lacks the density of real-world, production-ready implementation patterns needed to use it correctly. The Hacker News title frames the goal as moving agents off roughly 2019-era JavaScript and onto modern web primitives.

Modern Web Guidance works by having its CLI deliver focused guidelines, curated by platform experts, directly into the agent's context window. Three priorities shape the content: correct structuring of modern browser APIs that models frequently misuse; a preference for platform-level APIs that the browser can optimise and that carry built-in accessibility affordances; and lightweight fallbacks in place of heavyweight polyfills or legacy libraries.

What the guides cover

The project documents 250 modern web features and 203 real-world developer use cases. By the repository's own breakdown, 140 of those features concern CSS and layout, 36 cover HTML and DOM, and 74 fall under JavaScript and APIs.

Content is organised into six disciplines:

• User experience — view transitions, entry and exit animations, parallax scrolling, scrollbar-color
• CSS layout — container queries, subgrid, modern colour spaces such as oklch, text-wrap tuning, line-height trimming
• Performance — instant preloading, Interaction to Next Paint diagnostics, task scheduling with scheduler.yield
• Forms and UI — anchor positioning for tooltips, the Popover API, dialogs, :user-invalid validation, auto-sizing fields
• Accessibility — screen reader and keyboard operability, content navigation and discoverability
• Built-in AI — browser-native translation, summarisation and language detection

Because context window space is limited, the guides are written to be token-efficient, and the team runs evals to prune material that models already know, so tokens are spent only where guidance is genuinely needed. An eval dashboard and the full source code are published alongside the documentation.

Why it matters

AI assistants now write a large share of new code, and their habits compound: the patterns agents emit today become the baseline that code reviewers, refactor tools and future training data inherit. If agents trail the platform by several years — hand-rolling modal backdrops, shipping tooltip libraries or stacking polyfills where native primitives exist — that lag spreads through the ecosystem.

The project is also notable for who stands behind it. Rather than a single vendor promoting its own APIs, the content is jointly maintained by the teams behind Chrome, Edge and Firefox together with MDN, giving it cross-browser legitimacy that few comparable resources carry. It also models a pragmatic fix for stale model knowledge: instead of waiting for training corpora to catch up, current platform guidance ships as retrievable context, with evals keeping it lean.

For teams pointing agents at older codebases, the legacy-scan command doubles as a modernisation audit tool, and the repository is open for issues and community contributions.