What is Modern Web Guidance?
Modern Web Guidance is a set of agent skills plus a small command-line tool from the Google Chrome team. When a coding agent picks up a web task, the skill tells it to run `modern-web-guidance search`, which matches the request against a local index of guides, then `retrieve` to pull the chosen guide into context: code patterns, browser quirks and fallbacks for features such as the Popover API, anchor positioning, view transitions, container queries and `scheduler.yield()`. The README lists 262 web features and 204 use-case guides; the docs round this to 200+.
Version 1.0 shipped on 10 November 2026, after a preview at Google I/O in May, and 1.0.1 followed on 16 November. With 1.0, Firefox and Edge joined Chrome on the project, and the README credits the MDN team as a steward too. Recommendations carry Baseline browser-support data, a `legacy-scan` command looks for outdated patterns in an existing codebase, and an eval dashboard publishes pass rates with and without the skill. Code is Apache-2.0; guide text is CC BY 4.0.
Is Modern Web Guidance worth it?
Coding agents write web code from training data, and most of that data is old: a JavaScript modal where `<dialog>` would do, a scroll listener where a CSS scroll-driven animation would be lighter. This skill attacks that gap with retrieval rather than a long prompt. The skill file itself costs about 205 tokens; the guides only enter context when the agent searches for one, and search runs offline on an embedded model plus BM25, with no API key.
The evidence is better than most skills offer. The makers run each guide against tasks on two sample apps, with Playwright graders calibrated to pass every golden solution and fail the untouched app. In the 8 November snapshot in the README, Claude Code on Opus 5.5 passed 57% of 1,801 checks across 215 tasks without the skill and 88% with it, a gain of 31 points; Codex on GPT-6.1 went from 54% to 82%, and Antigravity on Gemini 4 from 58% to 94%. These are the project's own benchmarks on its own apps, so treat the size of the gain as an upper bound, but the method is published and the dashboard is public.
The guides also try to stop agents chasing novelty. Each one carries Baseline support data, so advice adapts to the browsers you target, and guides for Origin Trial APIs or features shipped in a single engine within the last 3 months are blocked at `retrieve` unless you opt in through a `.mwgrc` file. Fallbacks favour small custom code, under 50 lines, over heavy polyfills.
The catch: Google collects usage data unless you set `DISABLE_TELEMETRY=1`. That covers install counts, the calling agent's name, the guide IDs retrieved and the search queries the agent writes, though not your raw prompts. The `legacy-scan` modernisation workflow is deliberately narrow in 1.0, matching a limited set of patterns by static analysis. And if you installed through Anthropic's claude-plugins-official marketplace, your copy is stuck at v0.0.194 from 12 October; the README tells you to switch to the googlechrome marketplace.
Our take: for anyone who lets an agent write front-end code, this is closer to a style guide from the browser makers than to another prompt pack, and switching off telemetry is a one-line job.
Where it earns its place
A front-end developer on an agency team
Stop the agent building tooltips and modals from scratch in JavaScript
Install the skill and ask for the component as usual; the agent searches the guides, finds the Popover API or anchor positioning guide and writes native code with a fallback sized to your browser targets.
A tech lead on a five-year-old marketing site
Find polyfills and workarounds the browsers no longer need
Run `npx modern-web-guidance@latest legacy-scan .` for a static report, then ask the agent to modernise the flagged files with the matching guides. Coverage is limited in 1.0, so expect a partial list.
An accessibility specialist at a public-sector supplier
Get agent-written forms and dialogs right on focus, reduced motion and validation
The user-facing guides were reviewed with assistive technologies for 1.0; pairing them with a human audit catches the bad ARIA that the Chrome team says models still produce.
Before you commit
- Setup
- `npx modern-web-guidance@latest install` runs a wizard. In Claude Code: `/plugin marketplace add GoogleChrome/modern-web-guidance`, then `/plugin install modern-web-guidance@googlechrome`. Also installs through `npx skills add`, `gh skill install`, Codex, Copilot CLI and Antigravity.
- Integrations
- Claude Code, Codex, Cursor and other Agent Skills hosts, GitHub Copilot CLI, Antigravity, Grok Build CLI and Kimi Code. A second skill pack covers Chrome extensions; Chrome DevTools for agents pairs with it for live verification.
- Data and privacy
- Search and retrieval run locally with no network calls for content. Anonymous telemetry (install counts, agent name, guide IDs, search queries) goes to Google unless `DISABLE_TELEMETRY=1` is set; raw prompts are not collected.
- Cost at scale
- No paid tier. The cost is the tokens of the guides your agent retrieves; the base skill is about 205 tokens and the Chrome extensions pack about 175.
- Limits
- Origin Trial and volatile single-engine features stay locked until enabled in `.mwgrc`. legacy-scan uses static pattern matching only. Installs from claude-plugins-official are stranded at v0.0.194.
What’s good
- Maintained by Chrome with Edge, Firefox and MDN involvement
- Published evals with guided and unguided pass rates
- Offline search; guides load only when needed
- Baseline-aware advice, with experimental APIs gated
What’s not
- Usage telemetry on by default
- Benchmarks are run by the makers on their own sample apps
- legacy-scan covers a limited set of patterns in 1.0
- Copies from the official Claude marketplace no longer update
Best for
Front-end teams who let coding agents write HTML, CSS and JavaScript and want them to use current platform features with sensible browser fallbacks.
Modern Web Guidance alternatives
Modern Web Guidance: common questions
- Is Modern Web Guidance worth it?
- Modern Web Guidance is the rare agent skill backed by browser vendors and measured results, and the main thing to know before installing it is that telemetry is on by default. We rate it 4.4 out of 5.
- How much does Modern Web Guidance cost?
- Free and open source (Apache-2.0 code, CC BY 4.0 guides). Model usage is billed by your coding agent's provider.
- What are the downsides of Modern Web Guidance?
- Usage telemetry on by default; benchmarks are run by the makers on their own sample apps; legacy-scan covers a limited set of patterns in 1.0; copies from the official Claude marketplace no longer update.
- Who is Modern Web Guidance best for?
- Front-end teams who let coding agents write HTML, CSS and JavaScript and want them to use current platform features with sensible browser fallbacks.
- How much does Modern Web Guidance cost for a team or in production?
- No paid tier. The cost is the tokens of the guides your agent retrieves; the base skill is about 205 tokens and the Chrome extensions pack about 175.
- Is Modern Web Guidance safe for company data?
- Search and retrieval run locally with no network calls for content. Anonymous telemetry (install counts, agent name, guide IDs, search queries) goes to Google unless `DISABLE_TELEMETRY=1` is set; raw prompts are not collected.
- What are the best Modern Web Guidance alternatives?
- The closest alternatives we have reviewed are Agent Skills (Addy Osmani), Google Agent Skills and Anthropic Skills.
Answers checked Nov 20, 2026.