Index of every page on this site: llms.txt
All skills
emilkowalski avatar

/gestures

@64eb3de official
by Emil Kowalskiemilkowalski/skills58k stars
3,441
Run it yourself

Build drag, swipe, and sheet interactions that track the finger 1:1 and stay interruptible. Use when building or fixing anything the user drags, swipes, flings, holds, or dismisses with a gesture — drawers, sheets, swipe-to-dismiss toasts and cards, drag-to-reorder, pull-to-refresh, hold-to-confirm — or when a gesture feels laggy, disconnected from the finger, jumps on release, or can't be caught mid-animation. Triggers on — gesture, drag, swipe, fling, sheet, snap points, Vaul, Motion drag props, rubber-band, pull to refresh, hold to delete, touch-action, "doesn't follow my finger", "jumps when I let go", "can't grab it mid-animation".

Use this Skill: https://skilld.dev/gh/emilkowalski/skills/gestures

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈165 tokens for metadata: the name and description. ≈2k when used: this file.

Description uses 8.2% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈655 characters, or 8.2%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Gesture-Driven UI

How to build gestures that feel physical. A gesture is direct manipulation: while the finger is down, the UI is the finger; the moment it lifts, physics takes over. Every rule below follows from that split.

First decision: build it, or use a solution that exists?

Doing a drawer gesture right means momentum-based dragging, overlay opacity matched to drag progress, velocity-aware dismissal, focus trapping, and Escape-to-close. That is weeks of edge cases, not an afternoon. For a bottom sheet / drawer, default to Vaul — it exists precisely because this is hard to hand-roll. Build the gesture yourself only when no library covers the interaction (custom swipe cards, bespoke drag interactions). When unsure, read ../pick-ui-library/SKILL.md.

The two phases of every gesture

Walk this split for any gesture you build. Mixing the phases up is the #1 cause of gestures that feel wrong.

Is the pointer currently down?
├── Yes → TRACK phase: position maps 1:1 to the pointer.
│         No easing. No duration. No spring. The element IS the finger.
└── No  → RELEASE phase: physics finishes the motion.
          A spring seeded with the release velocity, or a CSS transition
          when the component is CSS-only (Vaul, hold-to-confirm). Never a keyframe.

Track phase rules

  • 1:1 means 1:1. Drive position from a useMotionValue (or direct transform writes), never a spring or a transition — a spring during tracking makes the element trail the finger and feel disconnected, like dragging something through syrup.
  • Write transform directly on the dragged element. Never update a CSS variable on a parent to move children: CSS variables are inheritable, so every change recalculates styles for all descendants. This exact bug made Vaul's drag laggy past ~20 list items; setting transform on the element fixed it.
  • No React state per move event. A re-render every pointermove drops frames. Motion values and direct style writes bypass the render cycle.
  • Rubber-band at the edges. Dragging past a boundary should move the element a fraction of the distance (Motion: dragElastic ≈ 0.2), not stop dead — hard stops feel like hitting a wall; resistance communicates "there's nothing further" physically.

Release phase rules

  • Custom drags finish with a spring, seeded with the gesture's velocity. The element should leave the finger at the speed it was moving. A fixed-duration ease here makes every fling land identically, which reads as fake.
  • Dismiss on distance OR velocity. Passing a distance threshold (~40–50% of travel) dismisses, but so does a fast flick over a short distance. In Motion, read both from onDragEnd's info.offset and info.velocity. Distance-only thresholds make flicks feel ignored.
  • Bounce belongs only at the end of a drag. A drag applies force — like throwing a ball at a wall — so a slight bounce on release feels physical. The same element closed by a button press gets zero bounce: no force was applied. Keep bounce at 0.1–0.3, and remember bounce scales inversely with element size (smaller elements need more bounce to read the same).
  • Snap-back is fast. If the gesture doesn't pass the threshold, return quickly with the spring — a slow snap-back feels like the UI is disagreeing with the user in slow motion.

Interruptibility is non-negotiable

The user must be able to catch a sheet mid-close and drag it back. Springs allow this: when retargeted mid-flight they keep their current velocity, so the redirect is seamless. CSS @keyframes restart from zero and cannot be caught — this is why Sonner's early CSS enter animation made toasts jump when a second toast arrived. Never use a keyframe animation on anything a gesture can touch. CSS transitions and springs only.

Linked values

The dragged element is rarely alone — an overlay fades, the page behind scales. Derive these from drag progress, don't animate them separately:

const y = useMotionValue(0);
const overlayOpacity = useTransform(y, [0, drawerHeight], [1, 0]);

One source of truth means the overlay is always exactly as faded as the drawer is dismissed — including mid-gesture, including when the user reverses. Two separate animations drift apart the moment a gesture is interrupted.

Motion drag props

const y = useMotionValue(0);

<motion.div
  drag="y"
  style={{ y }}
  dragConstraints={{ top: 0 }}
  dragElastic={0.2}
  dragMomentum={false}
  onDragEnd={(e, info) => {
    if (info.offset.y > threshold || info.velocity.y > 500) dismiss();
    else animate(y, 0, { type: 'spring', duration: 0.5, bounce: 0.2, velocity: info.velocity.y });
  }}
/>

dragMomentum is on by default — right for a flung card, wrong for a simple reposition; disable it when the element should stop where the finger stops. The example above disables it because onDragEnd hands the release to its own spring. dragConstraints takes a ref or bounds object.

Sheets without springs: the Vaul curve

Vaul mimics the iOS sheet with pure CSS to stay small: cubic-bezier(0.32, 0.72, 0, 1) over 500ms. The curve is extremely steep at the start — that's what lets a 500ms duration still feel snappy while landing with a spring-like gentle settle. Two rules from it:

  • A small floating drawer (not touching the screen edge) should override to ~200ms — the 500ms iOS timing only reads right on a full-width edge-attached sheet:
    [data-vaul-drawer]  { transition: transform 0.2s cubic-bezier(0.165, 0.84, 0.44, 1); }
    [data-vaul-overlay] { transition: opacity  0.2s cubic-bezier(0.165, 0.84, 0.44, 1); }
  • The component must feel like a single entity. If the drawer opens in 200ms, its height changes and content crossfades must live in the same timing family — one slow sub-animation makes the whole thing feel broken.

Mark interactive children with data-vaul-no-drag so pressing a button inside the drawer doesn't start a drag.

Spatial consistency

An element dismisses in the direction it entered. Sonner's toast slides up from the bottom, so swipe-down dismisses it — enter and exit share one axis, which is what makes the gesture feel discoverable and intuitive without instruction. A gesture that dismisses along an axis the element never moved on has to be learned; one that reverses the entrance is felt.

Hold gestures are asymmetric

Hold-to-delete and similar press-and-hold interactions use linear during the hold — the animation visualizes time passing, and time passes linearly; any other curve lies about how long is left. On release before completion, snap back fast with ease-out. The deliberate phase is slow and honest; the cancel is instant. Symmetric timing here is a bug.

Touch checklist

Run through this for every gesture you ship:

  • Test on a real phone — hit the dev server by local IP. Trackpad simulation lies about touch, and a 60Hz desktop monitor hides opacity-transition roughness that a 120Hz phone shows.
  • Hit targets ≥ 44×44px. Enlarge small visual targets with an invisible ::before hitbox instead of growing the layout.
  • touch-action on the drag axis (touch-action: pan-x for a vertical drag, etc.) so the browser doesn't fight the gesture with native scrolling.
  • Gate hover effects behind @media (hover: hover) and (pointer: fine) — touch fires phantom hovers on tap.

Reduced motion

Tracking is direct manipulation — the user is producing the motion — so it stays. What reduces: momentum flings, bounce, and any decorative motion around the gesture. On release under prefers-reduced-motion, settle with a short opacity/position change instead of a bouncy spring.

When it still feels wrong

A "weird" spring is usually fixed by increasing damping. A gesture that stutters is a performance problem — check the track-phase rules above first, then use the animation-performance skill. To diagnose a feeling you can't name, record the gesture and scrub it frame by frame.

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No alerts6d3 checks · Risk SAFE
  • Gen Agent Trust Hub6d

    This skill provides best practices and architectural guidance for implementing high-quality, interruptible gesture-based UI components. It focuses on performance, physics-based motion, and proper touch handling without introducing security risks.

  • Socket6d

    No alerts

  • Snyk6d

    Risk: LOW · No issues

Signed by skilld at 64eb3de. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 6 days ago.

Activeupdated last week

README badge

README badge for emilkowalski/skills/gestures