Twenty four automated checks failed. One passed.
i'm looking at a row on Chrome's public eval dashboard for Modern Web Guidance. The task is called Action menu: build a nav bar with nested submenus and on/off order options.
Unguided, the agent gets 1 of 25 checks right. Guided, it gets 25 of 25.
The failing version isn't broken, exactly. It works. It just does it the 2019 way.
Div overlays, a manual Escape keydown listener, z-index: 9999 .
Modern Web Guidance is a set of skills you install into your coding agent. Under the hood it's 200+ plain text guides about HTML, CSS, and JavaScript. When your agent picks up a frontend task, it runs a small CLI on your machine, searches those guides, pulls the good ones into context, and builds from there.
That's the whole trick. No fine-tuning. No new model.
A search command and a well-maintained folder.
Chrome shipped 1.0 on November 10, 2026, with Firefox, Edge, and MDN now contributing.…