Motion

Motion

Base invariants live in the active styling skill when one exists (prefer compositor-friendly properties, project motion tokens, and prefers-reduced-motion). In the minimal profile, apply those invariants using the repository's existing styling approach. This skill covers when and how.

Escalation ladder — use the cheapest tier that works

1. CSS transition — state changes (hover, open/close, theme). Default answer.
2. CSS keyframes — multi-step or looping effects.
3. Web Animations API — dynamic values or playback control from JS, still zero deps.
4. Animation library (Motion — formerly Framer Motion) — only for springs, gestures, layout/shared-element animation, or orchestrated sequences. It's a real dependency: the AGENTS.md ask-first rule applies; never add it for a fade.

Page & element transitions

• View Transitions API for page transitions (Astro: native cross-document view transitions, or <ClientRouter /> for SPA-mode extras; Next.js/SPA: progressive enhancement behind a document.startViewTransition feature check). It must be enhancement — navigation works identically without it.
• Verify page transitions with the browser Back button AND your E2E runner before shipping — two failure modes seen in practice: Astro <ClientRouter /> can drop page-scoped <style> on history back-navigation (the page reverts to an unstyled/full-width layout), and native @view-transition { navigation: auto } makes headless Playwright treat every link as never-"stable" so click() hangs. If a transition doesn't survive both, ship without it — the page must work identically anyway. Add a back-navigation smoke assertion to E2E (testing-playwright).
• Scroll-driven effects: CSS scroll-driven animations or IntersectionObserver. Never scroll event listeners doing style work — and if E2E must click elements inside a scroll-driven animation, prefer IntersectionObserver + a one-shot transition: scroll-linked animations count as permanently "running", which defeats Playwright's stability check.

Behavior rules

• Animations are interruptible: a second click mid-animation must not break state. CSS transitions handle this free; JS sequences must handle cancellation.
• Never gate interaction or content behind an entrance animation — the page is usable at frame 0, and LCP content is never delayed by animation. Hero/above-the-fold entrances: animate transform from a visible state, or animate secondary elements — never start the LCP element at opacity: 0 (the browser re-times LCP to when it becomes visible).
• Exit animations in React: the element must stay mounted until the animation ends — <AnimatePresence> with the library, or a two-phase state (closing → unmount on transitionend/animationend) by hand. Setting display: none in the same tick kills the animation.
• Durations from tokens; UI feedback stays fast (~100–250 ms); anything over ~400 ms needs a reason.

Reduced motion

• @media (prefers-reduced-motion: reduce): large/parallax/auto-playing motion is removed; motion that conveys state (a panel opening) is reduced to a quick fade or instant change — not deleted, the state change must still be perceivable. JS animations check matchMedia('(prefers-reduced-motion: reduce)').

Performance

• will-change only on elements about to animate, removed after — it's a per-element memory cost, not a go-faster switch.
• Avoid animating layout properties (width/height/top/left); for size transitions prefer transform: scale or FLIP, and interpolate-size/calc-size() where supported for height-auto transitions.
• Verify on a throttled CPU (DevTools 4x/6x) when browser tooling is available. Otherwise report HUMAN VERIFICATION REQUIRED rather than claiming the performance check passed.