motion-craft

name	motion-craft
description	Find, specify, review, or improve motion in a web interface against exact rules for frequency, purpose, easing, duration, choreography, interruption, gesture physics, accessibility, and performance. Use when the main task is animation, transitions, or interaction motion. Supports brief, opportunities, review, and improve modes. Do not use for reverse engineering a reference video or building an animated logo identity. Triggers on animation, motion, transition, easing, cubic-bezier, spring, stagger, drag, swipe, gesture, interruptible, prefers-reduced-motion, "make it feel better", "the animation feels off", "should this animate".
license	MIT

Motion Craft

Motion must clarify change, provide feedback, preserve continuity, or express a deliberate product character. Motion without a stated job is a candidate for deletion, not tuning. Default to flagging; approval is earned.

Most animation problems are not tuning problems. They are decisions that should have been "no".

Quick reference

Need	Read
Frequency gate, easing, duration, springs, choreography, performance, reduced motion	references/motion-principles.md
Drag, swipe, sheets, velocity, momentum, boundaries, gesture recognition	references/gesture-physics.md
Severity, findings table, rejected candidates, verdict	references/review-protocol.md
Frequency's role in product hierarchy, restraint, working order	references/interface-principles.md
Evidence boundaries and feel-checking	references/evidence-and-verification.md

Never approximate a value that exists in a reference. Copy the exact curve, duration, spring config, or threshold.

Select one mode

• brief — turn an idea into an exact motion specification.
• opportunities — find places that genuinely benefit from motion, and name what must stay static.
• review — assess existing motion without changing code.
• improve — produce prioritized, self-contained correction plans. Implement only after the user approves them.

Infer the mode only when the user's verb is unambiguous. Otherwise ask which result they want.

The ten non-negotiables

Every animation in scope is measured against these. A violation is a finding.

1. Justified. The purpose is one of: feedback, continuity, spatial consistency, state indication, preventing a jarring change, explanation, or delight. "It looks good" on a frequently-seen element is a block.
2. Frequency-appropriate. Keyboard-initiated and 100+/day actions get no animation. Tens/day gets ≤150ms opacity or color, or nothing. Occasional gets standard. Rare and first-run carry the delight budget.
3. Responsive easing. Entrances and exits use ease-out or a strong custom curve. ease-in on UI is a block. Built-in CSS easings are too weak for deliberate motion.
4. Sub-300ms UI. Anything slower on a UI element needs a stated reason. Per-element budgets are in references/motion-principles.md §4.
5. Physical origin. Trigger-anchored surfaces scale from their trigger, not center. Never scale(0) — enter from scale(0.90–0.97) with opacity. Modals are exempt and stay centered.
6. Interruptible. Rapidly triggered or gesture-driven motion retargets from its current value. Keyframes that restart from zero are a finding.
7. GPU-only. transform and opacity only. Layout properties, transition: all, and motion-library shorthand props under load are performance findings.
8. Accessible. prefers-reduced-motion honored — gentler, not zero. Hover motion gated behind @media (hover: hover) and (pointer: fine). Motion is never the only feedback channel.
9. Asymmetric where the interaction is. Deliberate phases animate slower; the system's response snaps. Symmetric timing on press-and-release is a finding.
10. Cohesive. Motion matches the component's personality and the rest of the product. Curves and durations live as shared tokens, not five hand-typed near-matches.

Flag on sight

• transition: all
• scale(0) entrances, or a pure fade with no initial transform
• ease-in on any UI interaction
• animation on a keyboard shortcut, command palette, or 100+/day action
• UI duration > 300ms with no stated reason
• transform-origin: center on a trigger-anchored popover, dropdown, or tooltip
• keyframes on toasts, toggles, or anything triggered rapidly
• animating width, height, margin, padding, top, or left
• motion-library x/y/scale shorthands on motion that runs while the page is busy
• a CSS variable on a parent driving a child's transform
• movement with no prefers-reduced-motion handling
• ungated :hover motion
• symmetric enter/exit timing on a press-and-release interaction
• an everything-at-once entrance where a stagger belongs
• drag dismissal on distance alone, with no velocity threshold
• a hard stop at a drag boundary instead of rising resistance

Shared intake

Establish before any mode proceeds:

• the user task and the interface state change;
• trigger, frequency tier, and input method;
• target elements and their spatial relationship;
• desired motion character and the product's personality;
• framework, installed motion libraries, and existing easing/duration tokens;
• browser, device, and performance constraints;
• reduced-motion requirement;
• reference material, and whether it is directional or pixel-exact.

Inspect existing motion tokens and component conventions before proposing new ones. Plans extend the project's system; they never add a parallel one.

Purpose gate

Complete this for every candidate before anything else:

Purpose:
Frequency tier:
User benefit:
Cost or risk:
Static / reduced-motion equivalent:
Decision: keep | reduce | remove | prototype

Reject motion that delays a frequent task, competes with content, makes position ambiguous, repeats without value, or duplicates feedback already communicated more clearly.

Brief mode

Produce, in order:

1. before and after state;
2. trigger and completion condition;
3. element hierarchy;
4. choreography phases;
5. per-property timing and easing, with exact values;
6. transform origin and spatial path;
7. interruption and reversal behavior;
8. responsive and reduced-motion variants;
9. implementation strategy;
10. verification cases, including the feel check.

Name the perceived quality alongside the numbers. A spec that reads 200ms ease-out without saying "crisp, arrives immediately" cannot be judged.

Opportunities mode

Inspect the actual journey. Expect to reject most candidates — an opportunity finder that suggests motion everywhere produces exactly the over-animated interfaces this skill exists to prevent.

Return three groups:

• high-value — motion measurably improves understanding or control;
• prototype — promising but needs comparison;
• keep static — motion would add cost, delay, noise, or accessibility risk.

Cap suggestions at 5–7 for a whole application, fewer for a single view, ordered by leverage rather than by how enjoyable they would be to build.

Where to hunt: pressable elements with no :active state; content that swaps, appears, or vanishes instantly; accordions that snap; panels and popovers with no spatial connection to their trigger; surfaces that exit differently than they entered; grids that pop in all at once; draggable elements that snap with no physics; rare high-emotion moments rendered flat.

Useful sweeps: conditional renders with no transition, onClick handlers on elements with no :active styling, details/accordion markup, drag handlers, .map( renders of entering lists, empty-state and success components.

A rejected-candidates section is required, each with the gate question that killed it. That section is what separates this from a wishlist.

Review mode

Observe at normal speed, at 10% speed, under repeated input, with the keyboard, and under reduced motion. Inspect the controller code separately from the visual result.

Review purpose and frequency; duration, easing, delay, and stagger; continuity and transform origin; gesture thresholds and velocity; interruption, reversal, and cleanup; layout, paint, and compositing cost; hidden-tab, offscreen, hover, and focus behavior; and reduced-motion behavior.

Report only proven findings, ordered by impact, using the table format in references/review-protocol.md. Cite file:line.

Improve mode

For every approved issue, write a plan that a model with zero context and zero taste can execute:

### NNN — <short imperative title>
- Severity / category:
- Current evidence: file:line plus the current code verbatim
- Desired behavior:
- Exact choreography: every value spelled out
- Files or components:
- Repo conventions to follow: one exemplar to imitate
- Steps: one concrete edit per step
- Boundaries: what must not be touched
- Accessibility:
- Performance:
- Mechanical verification: exact commands and expected outcome
- Feel check: what to watch for at 10% speed and under reduced motion
- Done when:

Never write "use the easing discussed above" — inline the exact cubic-bezier, the exact duration, the exact path. Do not bundle unrelated animation changes into one task.

Implementation selection

Choose the smallest tool that fits:

Situation	Tool
Bounded state motion, predetermined	CSS transitions
One-shot staged sequence	CSS keyframes
Entry without JavaScript	@starting-style
Programmatic control at CSS performance	Web Animations API
Stateful orchestration, layout continuity, exit presence	The project's existing motion library
Gestures, momentum, interruptible physics	Springs
Geometry-driven reveal or morph	SVG path, mask, or clip-path
A visual model DOM and SVG cannot express efficiently	Canvas or WebGL

Do not add a second animation library to solve a local effect without an explicit compatibility and maintenance justification. Check package.json and match the import path already in use.

Common mistakes

Mistake	Fix
Tuning an animation that should be deleted	Run the frequency gate first
transition: all	Name the properties
ease-in on a dropdown or modal	ease-out or var(--ease-out)
scale(0) entrance	scale(0.95) + opacity: 0
Popover scaling from its own center	transform-origin: var(--transform-origin)
Reporting a centered modal as a wrong origin	Modals are exempt
Keyframes on a rapidly triggered element	CSS transitions, so it retargets
Motion-library x/y shorthand under load	Full transform string
Reduced motion implemented as zero motion	Keep opacity and color, drop movement
Drag dismissal on distance only	Add a velocity threshold (~0.11 px/ms)
Five near-identical cubic-beziers	Consolidate into tokens
Approximating a value from memory	Copy it from the reference
Judging feel from source alone	Replay at 10% speed, or say it was not verified

Output

Follow references/review-protocol.md: a findings table with Severity | Location | Before | After | Why, a required rejected-candidates section, coverage, verification, and exactly one verdict — Block, Needs changes, or Approve.

Block on any feel-breaking regression: animation on a keyboard or high-frequency action, scale(0) or ease-in on UI, or a non-GPU animation with an easy GPU fix.

When feel cannot be judged from the available evidence, say so and put a feel check in the plan rather than guessing.