No-JavaScript Visibility Fix: test results

Verdicts

Date

2026-11-06

Strong · claude-sonnet-5-5 (Claude Code alias "sonnet")

Right on all 5 cases, read by hand. On a Next.js app it found the ready gate in the shared shell component, said that the use client mark on the page was not the cause, and covered every route; on a React Router site on static hosting it saw that prerender true leaves out the journal posts and listed them from the content file; on a plain Vite app it named the empty root, did not offer the noscript line as a fix and called copy in index.html a stop-gap for the home page; on an older Lovable project it said the host prerenders only for verified crawlers, sent the owner to the TanStack Start upgrade and told them not to add a prerender plugin; on a clean Astro docs site it answered exactly No findings.

Weak · claude-haiku-5-5 (Claude Code alias "haiku")

Right on 4 of 5 cases by the skill's own format checks, read by hand, but it missed the one-line format on the plain Vite case: the cause, the fix and the routes were correct, yet it added two extra lines after the Verify line, one listing what is not a finding and one listing what it could not check. It named the right layer on the Next.js, React Router and Lovable cases, kept the cookie banner and the tracking pixel out of every finding, and answered exactly No findings on the clean Astro site.

With and without the skill

Tested 2026-11-06.

Scored on content only. Read by hand, Sonnet without the skill found the ready gate in the Next.js shell and said the use client mark on the page does not empty the HTML, listed the journal slugs for prerender on the React Router site, named the empty root on the Vite app without making noscript the fix, and called the Astro site fine. It missed the Lovable case: it said Lovable cannot render on the server and advised a prerender plugin, with no word about prerendering for verified crawlers. Haiku did the same on all five, including the plugin advice. The skill added that one fact to each model, plus a fixed code per cause and the exact sentence for a clean input.

Same cases and the same checks with and without the skill. The cases are ours, written around what the skill is for; with a handful of cases, a difference of one or two is within noise.

Note

Five cases written by us (four planted, one clean): a Next.js app whose layout shell returns a spinner until an effect runs, a React Router site on static hosting whose prerender setting leaves out the dynamic journal pages, a plain Vite app with an empty root and a noscript line, an older Lovable project, and a clean Astro docs site. Each planted case gives only what the owner has before the diagnosis (a report line, the stack name and several real files, the guilty one among them with no comment pointing at it) and carries one trap and one safe look-alike (the use client mark on the page and a cookie banner loaded by script; a client-side search box; a tracking pixel in noscript and an analytics script; a chat widget and an analytics script). Each answer is scored by code for the skill's format, and the same answers are scored for content only (cause named, the fix avoids the trap and covers the real routes, a no-JavaScript check on a second route) so that the side without the skill can be compared. Facts were read on 6 November 2026 from the public pages of Google, Vercel, Next.js, React, Nuxt, Astro, React Router, Vite, Vike and Lovable; one fact the brief suggested, that a Vite SSR plugin was renamed, was not on any page read and was dropped. There are no owner-measured facts. One check was widened after the run for both sides: the forbidden-pattern check for copy put into noscript now needs copy, text or a headline to go into the tag, so an answer that only said to drop the noscript message is not flagged. One run per model and case.

What was not measured

• Models other than the two named above were not run.
• Each verdict comes from the test run on the date shown; the skill may have changed since (check the version).
• Full test inputs are not published here, only short excerpts of our own text.
• Results on your own texts, languages and domains can differ.