@mizchi/vlmkit-animation-eval

Frame-sampled animation evaluation for a web page. Every Web Animation on the page is paused and seeked through deterministic sample points; each sample is screenshotted and compared, and the report says whether the animation moves any pixels at all, when the page settles, whether prefers-reduced-motion is honoured behaviourally, and whether something outside the Web Animations API (rAF, video, GIF) is moving the page on its own.

It is the measurement behind two commands that live in different tools:

• vlmkit check animation page.html — the gate, with rule settings, --json, --strip filmstrips and the run ledger (package @mizchi/vlmkit-markup).
• vlmkit-anim eval page.html — the same report from the animation authoring tool, on the pages it emits (package @mizchi/vlmkit-anim).

import { runAnimationEval, formatAnimationEvalReport } from "@mizchi/vlmkit-animation-eval";

const report = await runAnimationEval({ source: "page.html", samples: 4 });
console.log(formatAnimationEvalReport(report));
report.issues; // [{ kind: "no-visible-effect" | "infinite-animation" | "reduced-motion-ignored" | "long-settle" | "uncontrolled-motion", severity, message, selector? }]

playwright is a required peer: the whole job is driving a browser. Depends on @mizchi/vlmkit-core for page loading, browser launch and PNG utilities, and on nothing else in the workspace.