Your Coding Agent Keeps Reaching for jQuery. This Project Aims to Fix That.

You've probably noticed it: ask a coding agent to build a modal or an animated transition, and it comes back with a pile of JavaScript and a legacy library or two. That's not the model being lazy—it's what most of its training data looks like. Modern Web Guidance is a project from the Chrome team (with help from Edge, Firefox, and MDN) that tries to steer agents toward the platform features that already do the job natively.

What It Does

Modern Web Guidance is a set of skills that embed web platform expertise, best practices, and browser compatibility data directly into your coding agents. The idea is straightforward: instead of hoping your agent remembers that a native API exists, you give it targeted, expert-curated guidance in its context window at the moment it's generating code.

The project ships as a CLI. You can search for relevant guides, retrieve a specific one by ID, or scan a codebase for legacy patterns worth modernizing. The guides themselves are designed to be token-efficient—the team runs evals to prune content that models already know, so you're not burning context on things the agent has already internalized. The source repository is where the authoring, calibration, and evaluation of those guides happens, and it's stewarded by the Chrome, Edge, Firefox, and MDN teams alongside the web development community.

Why It's Cool

• It targets the actual failure mode. The README is refreshingly blunt about why agents write legacy code: training data is full of it. Rather than trying to retrain anything, the project injects the missing knowledge at inference time. That's a pragmatic approach to a problem most people just complain about.

• The guides are opinionated about more than syntax. They cover modern browser APIs (helping models correctly structure APIs they frequently misuse), performance and accessibility (preferring platform-level APIs the browser can optimize, with built-in accessibility affordances), and responsible fallbacks. That last one matters—the guidance pushes toward lightweight fallbacks instead of heavy polyfills or legacy libraries, which is where a lot of agent-generated bloat comes from.

• Token efficiency is a design goal, not an afterthought. Anyone who's stuffed documentation into a context window knows the tradeoff. The eval-driven pruning means you get dense, practical guidance rather than a reference manual.

• You can try it without committing. The search, retrieve, and legacy-scan commands run straight from npx. You can see what the guides actually say before you install anything into your workflow.

• The governance is genuinely multi-vendor. Chrome, Edge, Firefox, and MDN all involved means this isn't one browser team's opinion about what "modern" means.

How to Try It

The fastest path is the interactive installer:

npx modern-web-guidance@latest install

If you'd rather kick the tires first, three commands let you explore without installing:

# Search for relevant guides
npx modern-web-guidance@latest search "animate a dialog modal backdrop"

# Retrieve a guide by ID
npx modern-web-guidance@latest retrieve "animate-to-from-top-layer"

# Scan a codebase for legacy patterns to modernize
npx modern-web-guidance@latest legacy-scan .

That legacy-scan command is a good place to start if you're curious about your own codebase—point it at a project and see what it flags. The full documentation lives at developer.chrome.com/docs/modern-web-guidance, and there's an eval dashboard if you want to see how the guides perform. The source is at github.com/googlechrome/modern-web-guidance-src, and issues go there too.

Final Thoughts

This is a tool for people who use coding agents on real web projects and are tired of deleting unnecessary dependencies. It won't make a bad agent good, but it addresses a specific, well-documented gap—the distance between "the model knows this API exists" and "the model knows how to use it well in production." The multi-vendor stewardship and the eval-driven approach suggest this is meant to be maintained, not just announced. If you're already running agents against a frontend codebase, the legacy-scan command alone is worth ten minutes of your afternoon.

Follow @githubprojects for more developer tools and open source projects.