The visual regression suite is twenty URLs

Your CI needs to catch CSS regressions on twenty key pages: landing, pricing, signup, checkout, eight product pages, the dashboard. The Percy bill came in and it was not small.

You write a shell script. It hits GET /screenshot?url=...&fullPage=true&format=png for each of the twenty pages, against a Hoody Browser pointed at your staging deploy. Twenty PNGs land in a directory. A pixel-diff against last week's baselines runs locally, posts the diffs to the PR.

The browser instance idles between runs. The container costs nothing while no CI is running. The next push wakes it up, twenty /screenshot calls later, the script either passes or hands you a PR comment with the rendered diffs.

Visual regression used to need a service. Now it needs a list of URLs.