The Best Animation and Motion Skills for Claude Code

We evaluated sixteen animation and motion skills for Claude Code. The scores span 9.7 down to 5.7, and the spread is not random — it maps exactly onto whether a skill knows which half of motion design it is supposed to be.

Motion is two problems stapled together

Ask an agent to "animate this" and you're really issuing two requests. The first is deterministic: easings, keyframes, timelines, timing curves — code with verifiable behavior. The second is judgment: does this feel right, does it serve the story, is 300ms too slow here? Agents are excellent at the first and notoriously unreliable at the second.

Most motion skills blur the two together. The HyperFrames family, from calesthio, is the highest-scoring body of work in the category precisely because it refuses to.

HyperFrames: the decomposition done right

The family's top scores are all on the deterministic side. hyperframes-cli at 9.7 is a rendering pipeline the agent can drive end-to-end. remotion-to-hyperframes, also 9.7, handles migration from the React-video ecosystem — a mechanical translation problem, and the skill treats it as one. hyperframes-core (9.5), hyperframes-media (9.5), and website-to-video (9.5) do asset handling and capture, equally mechanical, equally strong. The entry point hyperframes (8.9) and hyperframes-registry (8.7) route and discover.

The two most instructive entries sit on either side of the split. hyperframes-animation (9.4) encodes motion as atomic rules — timing, spacing, easing constraints an agent applies deterministically. hyperframes-creative (9.2) is the counterpart: creative direction, look, narrative arc — the judgment layer, explicitly separated rather than mixed in.

That architecture is the whole lesson. When the atomic-rules skill says "spring easing, 200ms, overshoot 6%," the agent executes verifiable instructions. When the creative skill says "make it feel like a documentary title sequence," it's framed as direction a human reviews, not behavior the agent invents. Neither skill asks the agent to be tasteful on its own. That's why the family holds 9.7 through 8.7 across nine skills while monolithic motion skills scatter.

The 3D bench

Three.js shows the same authorship effect we see across the directory. lhlGitHub's threejs-architecture-effects (9.5) and the standout threejs-world-generation (9.6) — the highest 3D score we've recorded — are tight, project-shaped skills with a clear definition of done. The gamedev-skills series is solid and consistent: threejs-materials-lighting at 9.2, threejs-gltf-loading at 8.9, threejs-scene-setup at 8.6.

Then calesthio — the same author dominating the video side — runs a flat line of 7.8 across threejs-shaders, threejs-loaders, threejs-lighting, threejs-interaction, and threejs-fundamentals. Not bad, but notably below their own HyperFrames work. The likely reason: shader and lighting craft resists reduction to atomic rules the way timeline code does not. The author's decomposition strategy, which wins at 9.5+ in motion, only gets them to the high sevens where the domain is genuinely judgment-heavy.

One-offs worth their slots

Outside the two families, four skills matter. anything2explainer (9.4) turns arbitrary input into motion-graphics-style explainer video — the rare high score for a creative-heavy skill, because it pins the style and leaves the agent to be mechanical. motion-graphics (9.0) does similar work as a general toolkit. vercel-labs' react-view-transitions (8.9) is the canonical take on the View Transitions API — small, official-adjacent, exactly scoped. And GSAP's own gsap-scrolltrigger (8.5) covers ScrollTrigger from the vendor, which once again beats the community alternatives on spec fidelity.

The cautionary tale at 5.7

svg-character-animation is the lowest-scoring skill in the category, and it fails in the exact way the pattern predicts. Character animation is almost pure judgment — squash-and-stretch, weight, anticipation — wrapped around a thin deterministic core of SVG transforms. The skill leans on the agent to supply the taste, gives it little in the way of rules, and scores collapse across our workflow and content dimensions as a result. It's not badly written; it's badly scoped. It asks an agent to do the one thing agents are worst at, with no structure to catch the failure.

The takeaway

The sixteen-skill spread in this category tells one story: motion skills succeed when they split deterministic code from aesthetic judgment, and route the judgment to a human with the rules made explicit. HyperFrames built an entire family on that split and owns the top of the table. Anything that asks the agent to freelance the aesthetics ends up at 5.7. Pick your motion skills on that axis before anything else — the full scores are in the Skill123 directory.