Testing Google’s “modern-web-guidance” skill against a real React app 0 ▲ Ahmad Alfy 43 minutes ago · 13 min read2688 words · Tech · hide · 0 comments LLM-assisted frontend work has a particular failure mode. The model confidently writes code that was best-practice in 2021. It reaches for 100vh, hand-rolls a dark-mode toggle with a class on <body>, or disables the submit button to “prevent” invalid input. None of it is wrong exactly. It’s just a few years stale, because the training data is a few years stale and the web platform moves faster than that. Google Chrome’s modern-web-guidance skill is a direct attempt to fix that. It’s not a linter and it’s not a codegen tool. It’s a search index over a curated set of best-practice guides, meant to be consulted before you write HTML/CSS/client-side JS, so the pattern you reach for is the current one. I wanted to know whether it actually earns its place in the loop. So I pointed it at a real codebase, the React frontend of a project-assessment internal tool I’ve been building, and treated it as an auditor. This is what came back. How the skill actually works There’s no magic. It’s two… No comments yet. Log in to reply on the Fediverse. Comments will appear here.
