Chrome's Modern Web Guidance Has 6.9k Stars and One Published Test: A Single Prompt, Run Twice
Software / news
Chrome's Modern Web Guidance Has 6.9k Stars and One Published Test: A Single Prompt, Run Twice
The agent skill lists 250 web features, and Chrome's own documentation page reports no eval results.

Chrome's Modern Web Guidance, a set of agent skills that tells coding agents which web platform features are safe to use, has reached 6.9k stars on GitHub, and the only head-to-head test of it we could find compares one prompt run once with and once without the skill.
The project lives in GoogleChrome/modern-web-guidance-src under the Apache-2.0 licence. Its README says it is stewarded by the Google Chrome, Microsoft Edge, Mozilla Firefox and MDN teams "alongside the web development community." The repository shows 1,612 commits, 559 forks, 305 open issues and 91 open pull requests.

What it contains
The README's reasoning is that agents "often default to older patterns because LLM training data contains vast amounts of legacy code." The guide index lists 250 web features and 203 developer use cases.
- CSS and Layout140 features
- JavaScript and APIs74 features
- HTML and DOM36 features
Source: GoogleChrome/modern-web-guidance-src README, accessed 2026-11-15
Installation is one command, npx modern-web-guidance@latest install, which opens an interactive wizard. Chrome's documentation page also lists Vercel's Agent Skills, a Claude Code plugin marketplace, Copilot CLI and Antigravity CLI. The CLI can search guides before anything is installed.
What has been measured
Chrome's documentation page makes no claim about evaluations or results, and it names no author. The README says the team runs evals "enabling us to prune lowest-common-denominator content that models already know" and links an eval dashboard, but no scores appear in the part of the README that was read.
The outside test is a LogRocket post dated July 22, 2026, by the writer Emmanuel John. He gave Gemini CLI one prompt twice, producing two Next.js task managers, and compared what the agent reached for each time.
| Feature | Without the skill | With the skill |
|---|---|---|
| Modal | Portal, a 300 ms setTimeout, manual body overflow | Native <dialog> with showModal() |
| Tooltip | @floating-ui/react | CSS anchor positioning |
| Cards | Viewport media queries | Container queries |
| Search over 2,000 tasks | useMemo filter | Batches of 50 with scheduler.yield() |
The differences are real, and so are the limits the author lists himself. He reports no bundle sizes, no INP or LCP figures and no frame timings, and each arm is a single run on a single agent. He also notes that CSS anchor positioning needs a fallback or a library for older browsers, and that native elements still need keyboard and screen-reader testing.
Who it changes work for
The task this changes is the first review of agent output. A developer who accepts an agent's @floating-ui/react import now has a competing diff that deletes a dependency, and a reviewer has to decide whether the browser support table allows the native version. The tool moves that check from the reviewer to the prompt, which helps only when the project's browser targets are written down.
The 305 open issues are the place to watch for whether guides disagree with each other or with real browser support. Alibaba's OpenCodeReview makes a similar bet on cutting agent cost, and Agweb is another browser-side tool for agents. LogRocket's post says Chrome calls Modern Web Guidance an early preview, so the first published eval numbers from the dashboard are the next thing that would settle the question.
Sources
More in Software
- 01Zigzag, a 35 MB Mac Browser, Courts Arc Users 35 Days Before ShutdownThe WebKit-based browser imports Arc spaces and pinned tabs, but its README lists no sync, no Boosts and a partial Chrome extension API.
- 02Nimony 0.6.2 Sets a Year-End Goal for Nim 3 With 44 Days LeftThe compiler that will become Nim 3 is MIT-licensed and builds natively without LLVM, but its FAQ says it cannot yet compile Nim 2 code unchanged.
- 03Alloy CSS Runs Menus and Tabs Without JavaScript, but Dialogs Need Safari 26.2The 24 KB MIT stylesheet leans on popover and invoker commands, so its no-script promise has browser floors that differ by component.
- 04Windows 11 Telemetry Flags Which Coding Agents You Run, Researcher FindsXusheng Li traced a service-experiment feature in build 26200.9974 that matches executable names such as claude.exe and queues the result as optional diagnostic data.