Back to Blog
What Modern Web Guidance 1.0 Changed in Praesepe LMS
I installed Modern Web Guidance 1.0 in Cursor on Praesepe LMS, a Vite and React app. Here is what landed in the repo, what the scanner found, and what I actually shipped.
Back
Engineering
Table of contents
Chrome shipped Modern Web Guidance 1.0 on November 10, 2026. It is a set of skills that steers a coding agent toward current HTML, CSS, and client-side JavaScript practice, with Baseline data and fallback advice in the guides. The npm package is Apache-2.0. The guide content itself is CC-BY 4.0.
I installed it in Cursor on Praesepe LMS, a pnpm monorepo with a Vite and React app in apps/web. I wanted to see what the 1.0 release actually puts in a real repo, and whether its new scanner would flag anything in CSS I already ship.
The skill tells the agent to call a local CLI: search to find a guide, then retrieve to pull the full text into context. Search runs offline, with no API key. You can also run those commands yourself, without installing anything.
Install
From the repo root:
bash
npx modern-web-guidance@latest install --agent cursor -yIt detected Cursor and installed only the modern-web-guidance pack at project scope. I did not need the chrome-extensions pack. It wrote .agents/skills/modern-web-guidance/, with SKILL.md and 209 guide files, plus skills-lock.json at the root. Chrome's 1.0 notes say 203 use-case guides; the extra files in the pack are broader discipline guides, such as accessibility and performance. It left the existing AGENTS.md and .cursor/rules alone.
One gotcha showed up right away: those official guide files would get sucked into our Prettier run. I added .agents to .prettierignore so Prettier leaves them alone.
The npm package wants Node 22 or newer. After install, the agent is still expected to call npx modern-web-guidance@latest when a task matches. Baseline defaults to Widely available unless you set a target, for example in AGENTS.md.
What legacy-scan found
Version 1.0 adds legacy-scan, which looks for a limited set of older patterns. There is no --fix flag. The loop is scan, verify, report, then modernize if you ask for that.
I ran it on apps/web. It checked 33 guides, about 20% of the catalog, and found two things. Both were the same pattern: stagger delays repeated with :nth-child, once in a skeleton list in motion.css, and again in tier animation lines in beacons.css. The suggestion was to replace those with CSS sibling-index(). That is also what the pack's own stagger-animation guide shows: one delay expression, instead of a rule per child. Both findings were already in CSS I ship, not something an agent had just written. That is the part I cared about: the scanner looking at an existing frontend, not a greenfield demo.
sibling-index() returns the element's 1-based position among its siblings, so you can write a single animation-delay: calc(sibling-index() * 0.1s) instead of a list of :nth-child rules. MDN marks it as Baseline 2026 Newly available since August 2026. I did apply a safe version of that change. The new delay uses sibling-index() behind an @supports check, and the old :nth-child rules stay in place for browsers that do not support it. I left the inline index * N delays in the TSX alone, because they were not repetitive, and in the account menu the separators would have thrown sibling-index() off.
What I actually shipped
The scanner only caught those two CSS hits. I went through the rest of the frontend by hand against the guides. Security headers, including a Content-Security-Policy in Report-Only, are now live on the server.
Here is what actually landed.
I split the app by route. When that change landed, the main entry went from 283.69 kB gzip to 154.62 kB gzip. I have not remeasured a later build.
The QRIS payment countdown no longer talks to a screen reader every second. It announces only at milestones.
There is a skip link to main content.
The account menu is a disclosure popover now. I kept the old menu as a fallback.
The useful part, for me, was not the install itself. It was a concrete list from an app I already maintain, then a set of changes I could ship with fallbacks still in place.
Sources
Comments
No comments yet
Loading comments...