Turn motion code, recordings, and traces into a release verdict with prioritized fixes for timing, interruption, reduced motion, and runtime cost.
An animation can look fine once and still fail the real interaction. A second click restarts it. A route change leaves stale state behind. Reduced motion removes the only feedback. A clean code diff drops frames on the device that matters.
This skill reviews motion as product behavior. It inventories each transition, separates code, recording, runtime, and team evidence, and checks purpose, frequency, timing, spatial logic, interruption, accessibility, performance, and token consistency. Each finding leads to a keep, tune, replace, reduce, remove, measure, or block decision with an owner and acceptance check.
Use it before shipping web or app motion, when a component feels wrong, or when design and engineering disagree about readiness. The package includes a detailed review rubric, a canonical JSON schema, and a read-only checker for cross-references, blocker consistency, reduced-motion coverage, and release gates. It does not watch an interface you have not supplied, invent profiler results, or treat a preferred curve as proof of runtime quality.
Input
Review this React and Framer Motion diff before release. The account menu scales in over 360 ms. Toasts use keyframes and restart when another save finishes. Product says both feel polished. We support keyboard and pointer input. No device profile or reduced-motion test is attached. Do not edit the code.
Output
A conditional or blocked verdict; separate code observations from missing runtime evidence; motion inventory rows for menu entry, dismissal, and repeated toasts; findings for delayed feedback, restart behavior, focus and exit state, reduced-motion coverage, and unmeasured frame delivery; bounded fixes; and keyboard, rapid-repeat, interruption, reduced-motion, and target-device gates.
Input
Provide the component or product surface, target platforms and inputs, the release or diff, and the evidence you have: source, motion tokens, design specs, recordings, traces, profiler captures, accessibility tests, or bug reports. Name the review mode and whether changes are allowed. The skill returns a scoped verdict, evidence register, motion inventory, prioritized findings, bounded remediation specs, and acceptance gates without pretending unrun checks passed.
Source code, a diff, prototype, design specification, or component contract that identifies the states and transitions to review.
Platforms, browsers or OS versions, devices, input methods, user actions, frequency, and release scope. Unknown context can remain explicit and may produce a measure-first verdict.
Recordings, runtime profiles, accessibility checks, test results, or issue reports when available. Code alone cannot prove delivered frames or complete device coverage.
Optional. Required only to run the bundled read-only canonical record checker. It uses the standard library and does not need network access.
SKILL.md; UI motion evidence, purpose, timing, state, accessibility, performance, system, severity, and verdict rubric; human report template; canonical motion-review JSON schema; read-only record checker; Agent Skills interface metadata
No reviews yet.
Our SwiftUI checkout zooms the selected card to full screen and parallax-scrolls the confirmation. With Reduce Motion enabled, the app disables every transition, including the success-state change, so users sometimes tap Pay again. A screen recording from one iPhone and the relevant view code are attached. Decide what blocks release.
Output
An evidence-bound blocked verdict; separate card, payment-progress, success, and confirmation motion instances; a critical or high finding for lost payment feedback; a calmer non-spatial alternative that preserves state; repeated-tap, VoiceOver, Reduce Motion, backgrounding, stale-completion, and payment-state checks; and no claim that one recording proves all supported devices.
Input
A dashboard drawer animates only transform and opacity, and a 60 fps desktop recording looks clean. One Android trace on build 842 shows two long tasks and missed frames while charts mount. The drawer code, trace notes, device model, and motion tokens are supplied. The team wants approval because the CSS uses compositor-friendly properties.
Output
A ready-with-conditions or blocked verdict scoped to build 842 and the tested device; code, recording, and trace evidence kept separate; a causal hypothesis around chart work rather than an automatic CSS pass; contained tests for drawer-only and chart-mount variants; input, focus, interruption, reduced-motion, frame, long-task, memory, and rollback gates; and no invented performance result.
Creator
Tterivo5
