Repository navigation
feat(sandbox): port the player.style instaplay theme to v10 elements - #2714
cjpillsbury wants to merge 23 commits into
Conversation
Adds a sandbox page that renders player.style's instaplay theme on Video.js 10 elements, as a first probe of what porting the player.style catalogue costs. The theme ships as markup plus a stylesheet and is stamped into an element by the sandbox's existing `defineTemplateSkin`, the same path an ejected registry skin takes. No packaged-skin API is involved, so nothing here depends on how third-party skins eventually get distributed. Translation notes: - Named icon slots become children toggled by CSS on the state data attributes. `data-volume-level` reports off/low/medium/high and the icon set ships three glyphs, so medium borrows the high one. - `media-time-range` becomes `media-time-slider` with explicit track, buffer and fill children, driven by `--media-slider-fill` and `--media-slider-buffer`. - The theme marks its bar and range `noautohide`, so they sit outside `media-controls` rather than configuring autohide. - The skin brings its own box, since the packaged container's sizing lives in a stylesheet this page does not load. Also declares `*.html?raw` in the sandbox ambient types, which the existing registry-skin imports were already relying on without one. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
✅ Deploy Preview for vjs10-site ready!
To edit notification comments on pull requests, go to your Netlify project configuration. |
✅ Deploy Preview for vjs-registry canceled.
|
@videojs/cdn
@videojs/core
@videojs/element
@videojs/html
@videojs/media
@videojs/react
@videojs/spf
@videojs/store
@videojs/utils
@videojs/cloudflare-video
@videojs/dash-video
@videojs/hlsjs-video
@videojs/mux-audio
@videojs/mux-video
@videojs/native-hls-video
@videojs/shaka-video
@videojs/spotify-audio
@videojs/tiktok-video
@videojs/twitch-video
@videojs/vimeo-video
@videojs/wistia-video
@videojs/youtube-video
@videojs/google-cast
@videojs/mux-data
commit: |
📦 Bundle Size Report🎨 @videojs/html — 34 small size changes
Presets (7)
Media (18)
Players (5)
Skins (29)
UI Components (62)
⚛️ @videojs/react
Small changes (40, ≤ 300 B)
Presets (7)
Media (22)
Extensions (2)
Players (5)
Skins (18)
UI Components (39)
🧩 @videojs/core — 1 small size change
Entries (76)
🏷️ @videojs/element — no changesEntries (2)
📦 @videojs/store — no changesEntries (3)
🔧 @videojs/utils — no changesEntries (13)
📦 @videojs/cdn — no changes📦 @videojs/cloudflare-video — no changes📦 @videojs/dash-video — no changes📦 @videojs/google-cast — no changes📦 @videojs/hlsjs-video — no changes📦 @videojs/media — no changesEntries (3)
📦 @videojs/mux
📦 @videojs/mux-audio
Entries (2)
📦 @videojs/mux-data — no changes📦 @videojs/mux-video
Small changes (1, ≤ 300 B)
Entries (2)
📦 @videojs/native-hls-video — no changes📦 @videojs/shaka-video — no changes📦 @videojs/spf — no changesEntries (7)
📦 @videojs/spotify-audio — no changes📦 @videojs/tiktok-video — no changes📦 @videojs/twitch-video — no changes📦 @videojs/vimeo-video — no changes📦 @videojs/wistia-video — no changesEntries (2)
📦 @videojs/youtube-video — no changesℹ️ How to interpretEach entry is independently bundled, minified, and brotli-compressed. Initial size includes its static import graph; lazy dynamic chunks are reported separately. Entries are not additive because their dependency graphs overlap. Preset rows represent realistic combined bundles. Changes of 300 B or less across initial, lazy, and total size are collapsed, not discarded. Run |
…iginal Splits the page into three modes on one route. Without `panel` it is a two-frame comparison; with it the page is a single player, which is what each frame loads. The comparison follows the shell's Compare: one axis differs, everything else is shared, each panel is its own frame, and a checkbox mirrors playback between them. The axis here is the port itself, which cannot live in the shell's Compare because every axis there differs on a shell-owned selection. Two frames is a requirement, not a preference. media-chrome and `@videojs/html` register eleven of the same custom element names and whichever loads first silently wins, so each stack gets its own module and its own document. The reference panel loads the published theme from a pinned CDN bundle rather than a dependency, so the StackBlitz upload does not have to resolve it. The bundle carries its own pinned media-chrome and never reaches the workspace copy. `installSandboxMirror` needed no changes: it moves state through the media element, so it mirrors across two different player stacks as-is. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
… page The comparison shell, both panels and the router were written inside the instaplay template. Thirteen more themes are coming, so they move to `app/shared/player-style/` and a theme becomes markup, a stylesheet and two small entries. `themes.ts` is the registry: every theme's preset and the published version its reference frame pins. Pinned rather than a range, since bumping one changes what the comparison compares against. Each panel entry now imports `@app/styles.css` itself instead of relying on a caller, and `elements.ts` registers the presets alongside the UI elements. Both were latent in the single-page version, where one entry happened to import everything. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Second theme on the shared harness, with its comparison frame beside it. Icons are the original's own SVGs rather than v10's set, since the theme ships them and they carry its look. Two things the side-by-side surfaced, both of which will repeat: - v10 splits what media-chrome collapses. A control whose feature the platform lacks entirely reports `data-hidden`; one whose API exists but has nothing to offer — Chrome with no Cast device — reports `data-availability="unavailable"` and stays interactive-looking. media-chrome drops the control in both cases, so a faithful port hides both. - The `sm:600` breakpoint becomes a container query on the skin root, which is how the packaged v10 skins handle the same thing. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Halloween is the interesting one. The original reaches into media-chrome's shadow DOM with `::part(track)` and `::part(thumb)`, and stacks a second range element because a shadow part cannot carry an animation from the theme. v10 builds its slider from real children, so the parts become ordinary selectors and the duplicate element disappears — the port is smaller than the original. Two more differences that will repeat: - media-chrome paints button icons from `--media-icon-color`, which overrides the artwork's own `fill`. v10 leaves icon paint to the skin, so artwork that ships `fill="none"` and expects the player to colour it renders blank. The halloween pumpkin lost its carved face to this. - v10 reports volume level on `media-mute-button` only. Halloween styles its candle by level but has no mute button, so the port carries a hidden one as a state carrier. Notflix steps `--media-font-size` at four breakpoints; those become container queries at the same stops. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
First theme with binary assets. The original loads its pixel art through player.style's build step, which inlines each PNG with an ejs `base64()` call; a v10 skin is static markup with no such step, so the fourteen PNGs are inlined at author time instead. Any real port needs an equivalent build step. One detail does not carry over. The original nudges its slider thumbs a pixel or two with selectors like `media-time-range[mediacurrenttime^="0"]`, reading media-chrome's reflected state attributes. v10 exposes position as CSS variables and reflects no numeric state, so those sub-pixel corrections are dropped. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
First audio-preset port, and the one that exercised the icon pipeline hardest. The original is written in Tailwind utilities and compiles its own stylesheet. The sandbox already scans `templates/`, so the port keeps the utilities inline and needs almost no authored CSS. Three things this theme surfaced, all of which apply beyond it: - Its icons are `<use href="#id">` against a sprite defined at the top of the template, outside the controller. The sprite has to come across with the markup or every control renders an empty box of the right size. - media-chrome sizes button icons for you through `--media-button-icon-height`. v10 leaves sizing to the skin, so icons need explicit dimensions or they collapse. - v10's playback-rate button carries no text of its own; it reports `data-rate` and the skin draws the label. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Adds a theme picker to the comparison header so themes can be swapped without navigating, keeping every other selection in the query string — the way the shell's own pickers work. Also writes down what had been living in my head: which themes are ported, what is still off in each, and — importantly — that "checked" so far means one width, one source, paused, no captions, no menu open. A blank deltas cell means not looked at, not matches. Corrects two things earlier commit messages overstated: - Icon colour is not missing from v10. Its icons ship `fill="currentColor"` and follow CSS `color`. What differs is that media-chrome forces `fill` onto slotted artwork and v10 never touches skin-authored SVG. - Icon sizing is not absent either. `--media-icon-size` exists and the packaged skins size through it; it is marked internal, so a skin outside the package supplies its own. Records one real upstream bug: the five availability-gated buttons compute `hidden` four different ways, and cast is the outlier that leaves a dead control visible in the packaged skin. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Artwork is referenced from player.style over a pinned CDN rather than copied in. That theme is where these assets live, and the media-chrome repo already does the same thing — a one-kilobyte example that consumes the published theme rather than carrying the bitmaps. This is the theme whose transport sits entirely outside `media-controller`, bound with `mediacontroller="controller"`. v10 resolves state through DOM ancestry with no id binding, so the controls are simply descendants of the player and `media-container` wraps only the video region. The structure ports unchanged. Also fixes the theme picker, which showed the wrong label: `selected` set on a detached option does not survive being appended, so the value is assigned after insertion. And repairs `PORTED_THEMES`, which had silently lost two entries — the formatter reflowed the array across lines and later single-line edits no-opped against it. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Structurally the same shape as demuxed-2022 — bottom scrim, decorated top chrome, centred play, a control bar with the slider taking the remaining width. The artwork is inline SVG in the original template rather than separate asset files, so it stays inline here; that is most of the theme's size. Its `sm:600` breakpoint, which only shrinks the slider thumb, becomes a container query at the same stop. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The clearest case of a theme that delegates rather than styles. Sutro's own stylesheet is 66 lines: it sets a `--base` scale and a handful of `--media-*` variables and lets media-chrome's built-in control styling do the rest. v10 has no equivalents for most of those, so the port authors what the original got for free — control sizing, the menu surface, every slider part. Its `--base` steps 18 → 20 → 24px across a width breakpoint and fullscreen, which is what `--media-scale-unit` does for the packaged v10 skins. That variable is internal, so the scale stays local to the theme. Also the first port with a nested settings menu: media-chrome's single element with `slot="submenu"` children becomes `media-menu` plus sibling `media-menu-content` panels addressed by `commandfor`, each list supplied by a radio group rendering its `<template>` per option. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Summary
Failed runRun 34612404353 · head SHA Evidence
Likely causeVerified: the failure happened inside the generated Next/Tailwind consumer's build before browser coverage began. Inference: this looks like a toolchain/runtime failure in that build path, not evidence of a product regression from this sandbox-only diff; the available output does not identify the crashing binary or a reproducible source-level cause. Recommended actions
|
Second audio preset, and the same delegate-heavy shape as sutro: 71 lines of its own CSS, everything else inherited from media-chrome's control styling, so the port authors the sizing and slider parts off the `--base` scale itself. Its `mediatitle` and `mediabyline` template expressions become slots the page fills, which is the shape the packaged skins already use for a replaceable poster. Also widens the harness poster resolver. It only looked for `media-poster img`, which video themes use; audio themes place the artwork themselves behind a named slot, so theirs was being dropped and the cover never appeared. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Three separate menus rather than one settings menu, so each gets its own `media-menu` and trigger. The original's `bounds="vimeonova-bounds"` — media-chrome's way of constraining the preview box to an element — becomes the slider preview's own `overflow="clamp"`. Records a harness limitation the theme exposed: skins are stamped into light DOM, where `<slot>` is inert. Themes that expose a title or byline through a named slot render nothing, and an empty chip where the markup wraps it. Only a shadow-DOM skin — v10's own `SkinElement` — can honour named slots, which is a point in favour of that being the eventual distribution shape. Hidden for now in vimeonova and sutro-audio, which both have one. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Same settings-menu shape as sutro, plus a set of centred controls for narrow widths. Another delegate-heavy theme — 57 lines of its own CSS — so the port authors the control sizing and slider parts itself. Its largest size is tied to fullscreen rather than a width, which ports directly as `:fullscreen` on the skin root. Two deltas recorded rather than chased, per the first-pass standard: the bar's play glyph renders as a skip icon, and the centred seek controls appear at the wrong breakpoint. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The theme the audit called hardest, and the reason is composition rather than size. media-chrome let it render a different *set* of controls per breakpoint through `<template if="breakpointsm">`. v10 skins are static, so every control is always in the tree and container queries decide what shows. Hidden controls use `display: none` so they leave the tab order and the accessibility tree rather than just the picture. Only the on-demand half. The original also branches on `streamtype == 'live'`, which in v10 is a separate preset and therefore a second skin the harness does not yet mount. Known delta: at a given width the port shows more controls than the original, so the thresholds still need reconciling against media-chrome's breakpoints. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Last of the fourteen. Its controls are a centred cluster rather than a bottom bar, with the scrubber directly underneath. On-demand only. The original branches on `streamtype == 'live'` and again on whether there is a target live window, which in v10 means the `live-video` preset and a second skin. Known deltas: the cluster sits centre-vertically where the original sits it near the bottom, and more controls show at a given width than the original's breakpoints allow. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
SummaryClassification: inconclusive. The registry-consumer suite failed during global setup; no Playwright test began. Failed run
Evidence
Likely causeVerified: the Recommended actions
|
…nd yt All three were recorded as breakpoint-threshold mismatches. That diagnosis was wrong; none of them was a threshold problem. - `minimal` and `microvideo` default their seek and picture-in-picture buttons off behind a custom property, so an integrator opts in. Three rules near the end of each stylesheet, which I missed by reading only the first sixty lines. - `yt`'s play artwork ships both states as sub-groups of one SVG with no rules of its own; media-chrome toggled them from the player's stylesheet, so both painted at once and overlapped into an unreadable glyph. - `yt`'s centred seek controls are not width-gated in the original — that gate was mine — and they render empty because media-chrome supplied default icons for an empty button where v10 supplies none. Also drops the duration from `minimal`'s readout, which shows elapsed time only, and bottom-aligns `microvideo`'s cluster with the scrubber pinned to the edge, matching where media-chrome's centred-chrome grid puts them. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Completes the catalogue at sixteen skins from fourteen themes. media-chrome selected the live branch with `streamtype == 'live'`; v10 puts live on its own preset, so each is a second skin compared against the same published package. The harness gains live support: `mountPortedPanel` now takes the player tag from the sandbox context rather than hardcoding it, so the live preset is picked up, and a live port mounts `hls-video` — a plain `<video>` cannot play the live source in Chrome, and without playback the stream type never resolves and the live controls stay inert. The comparison lands live themes on a live source when the URL names none, since comparing a live skin against an on-demand source shows neither side's live branch. Each live skin imports its on-demand sibling's stylesheet and adds only what the live branch needs, so the two stay in step. Known simplification: the originals branch again on whether there is a target live window, showing a scrubber and seek controls only with DVR. v10 exposes no equivalent to select on, so these render the DVR shape and the no-DVR variant is unported. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The live ports mounted `<live-video-player>`, but `elements.ts` registered only the video and audio presets — so it was an unknown element with no store, and no state reached any control. The media played, because `hls-video` plays independently of the player, and the skin rendered, so it looked correct while the live button, scrubber and everything else were inert. Verified: the live button now reports `data-live` and the slider receives a fill value. Also corrects the note on DVR. v10 does model it — `targetLiveWindow` is in store state and the live button already reads it. What is missing is a data attribute reflecting it, which is what a static skin needs to select on. v10's own live skins avoid the question by carrying no scrubber at all. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Adds a React implementation of the instaplay port beside the hand-written HTML one, reached with `?impl=react` and toggled by a checkbox in the comparison header. The reference frame is untouched either way, so both implementations are compared against the same published original. The skin sizes its controls in `em` and steps each one to 17px once the container reaches 384px, which is how the original scales. The step sits on the controls rather than the container: a container query matches ancestors only, so the element declaring the container cannot answer its own query — the original bumps `[role='button']` for the same reason. Two things the React target leaves to the skin that media-chrome does for you: a `<video>` the page authored needs sizing, or it lays out at its intrinsic dimensions inside the skin's box, and `Poster.Root` reports visibility through `data-visible` without acting on it. Also moves the theme's source default into `applyThemeDefaults`, shared by the comparison shell and the panel route. It used to live in the shell alone, so a live theme's panel URL opened on its own fell back to the sandbox's on-demand default and reported the source as unplayable. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Every port, plus the React implementation, loaded and checked for the things only the rendered page can answer: the media fills the skin's box, a play control exists and survives Playwright's actionability check, clicking it reaches playback and the clock moves, and no image is left painting over the video once it plays. Each check is pinned to a defect this suite was written against. Player state alone reported all of them as working: a video left unsized plays perfectly while laying out at its intrinsic size, a poster that never hides covers a video that is playing behind it, and a control under an overlay reports the right state while swallowing the click. The pages load with `preload: none` deliberately. Loaded media hides the sizing defect rather than exposing it — a frame wider than the skin is capped by the page's `max-width` reset and lands on the right size for the wrong reason, while a narrower one stays small. With nothing loaded there is no intrinsic size, so the box is the skin's CSS and nothing else. The published live themes carry no play button, so their ports have none either; those reach playback through the media element instead. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Interactive Playwright runs against the published original, state by state, turned up four things the fixed-state checks never saw. The progress bar never rendered. `Slider.Fill` and `Slider.Buffer` render a bare element and report progress only through `--media-slider-fill` and `--media-slider-buffer`; a layer that reads neither has no extent, so both sat at zero width through every state while the variables read correctly. They now clip to their variable, which is what the packaged skins' own `clip-media-x-*` utility does, and the fill swaps to `--media-slider-pointer` while dragging so it follows the pointer rather than the committed time. The preview had no thumbnail. The panel never passed the storyboard track the hand-written port gets from the same helpers, and the thumbnail carried no size cap — `Thumbnail.Root` scales a tile to whatever its computed style allows, so an uncapped one rendered half again as wide as the original's. The cap is declared on the preview and read by the thumbnail, as the packaged skins do, at the 180px the original uses. `Controls.Content` covered the whole surface, so it sat above the centred play button and the gesture layer and had to be excused with `pointer-events`. The packaged skins position their control groups instead; anchored to the bottom edge and only as tall as its two rows, nothing needs excusing. The container now carries `isolate` and names itself `media-root` after the packaged container, and the media and poster take that container's rule for slotted media, down to reading the public `--media-object-fit`. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The fill sat at zero width in the React port through every state the suite already checked, because nothing in it looked at the progress bar. This compares the fill's rendered extent against the share of the track `--media-slider-fill` claims. The extent is read back rather than the declaration: a skin reads the variable as a width or as a clip on a full-size layer, and either way a layer that reads nothing renders at zero however plausible its markup looks. Both forms name the variable in the declaration that uses it, so one selector finds either. Where a theme has a volume slider too, the fills are indistinguishable by markup and the widest track wins. Polled rather than read once, since several of these skins transition the fill. Winamp opts out: it shows position with a thumb on a bare track and has no fill layer, as the original does. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
SummaryClassification: inconclusive. The Registry consumers job failed before any Playwright browser test ran: TypeScript aborted with a native stack overflow. Failed runRun 34657478266 · head Affected tests: none executed; failure occurred in Registry consumers global setup while verifying generated React consumers. Evidence
Likely causeVerified: the failing process is the TypeScript compiler invoked by Inference: this is a TypeScript/toolchain stack-overflow failure or an interaction in the generated consumer environment, rather than evidence of a regression introduced by the sandbox-only diff. Recommended actions
|
|
used as starting point / superceded by efforts from @decepulis |
Draft — exploratory, not proposing a merge yet. Opened so the diff is easy to audit.
What this is
A sandbox page at
/player-style-instaplay/that renders player.style'sinstaplaytheme rebuilt on Video.js 10 elements, side by side with the published media-chrome original. It's the first probe in an audit of what porting the whole player.style catalogue (14 themes) would cost.The page
Three modes on one route:
/player-style-instaplay/?panel=videojs?panel=media-chromeThe comparison follows the shell's Compare: one axis differs, everything else is shared, each panel is its own frame, and Mirror playback carries play/pause, seeks, volume, mute and caption selection between them. Layout and mirror travel in the URL, so a link reproduces the view.
The axis here — the port itself — can't live in the shell's Compare, whose axes all differ on a shell-owned selection (
styling,skins,skin,platform).Two frames is a requirement, not a preference: media-chrome and
@videojs/htmlregister eleven of the same custom element names (media-play-button,media-mute-button,media-container,media-fullscreen-button,media-pip-button,media-airplay-button,media-cast-button,media-captions-button,media-live-button,media-playback-rate-button,media-tooltip). Both sides guard registration and silently no-op, so whichever loads first wins and the other's controls render as inert boxes. Each stack therefore gets its own module and its own document.installSandboxMirrorneeded no changes — it moves state through the media element, so it mirrors across two different player stacks as-is.How the port is built
theme.html+theme.css, stamped into an element by the sandbox's existingdefineTemplateSkin— the same path an ejected registry skin takes, on the publicContainerElement. No packaged-skin API is involved, so nothing here depends on how third-party skins eventually get distributed. That question is open and this PR deliberately avoids it.The reference panel loads
@player.style/instaplay@0.1.2from a pinned jsDelivr+esmbundle rather than a dependency, so the StackBlitz upload doesn't have to resolve it. The bundle carries its own pinned media-chrome and never reaches the workspace copy.Translation notes
slot="play",slot="off", …)media-time-rangemedia-time-slider+ explicit track / buffer / fillmedia-preview-time-displaymedia-slider-value type="pointer"inmedia-slider-previewmedia-preview-thumbnailmedia-slider-thumbnailmedia-controllervideo-player+media-containerFour things worth a look, because they'll repeat across every other theme:
media-controls. instaplay marks its bar and rangenoautohide, so they sit in a plain always-visible wrapper.ControlsElementonly consumesplayerContext, so nothing breaks — but ported themes opt out of autohide entirely rather than configuring it.width: 100%; aspect-ratio: 16/9). The packaged container'sw-full h-fulllives in a stylesheet this page doesn't load; without it the player is 0×0. The reference panel needsw-fullfor the same reason —aspect-videoandmax-w-*never set a width, and a theme whose media hasn't loaded has nothing to size itself from.data-volume-levelreportsoff|low|medium|high; the icon set ships off/low/high, so medium borrows the high glyph.Also included
declare module '*.html?raw'in the sandbox ambient types. There wasn't one, so the existing registry-skin?rawimports were untyped too.Verified
Both panels render side by side and read as equivalent. Port renders at 896×504 inside
<video-player>; poster, centred play button and mute control correct;data-pausedshows/hides the play button; the poster hides on play;data-volume-levelswaps the mute glyph; the time slider sits flush at the bottom edge with v10 writing--media-slider-filland--media-slider-buffer. Mirror confirmed end to end — muting and setting volume in the port propagated to the original through the relay. Clean console in all three modes. Typecheck adds no errors.Not verified: playback. Media doesn't load in my browser —
readyState: 0,duration: NaN, no video request while poster and storyboard fetch fine. The stockhtml-videotemplate shows the identical state with the same source, so it's environmental rather than something this branch introduced. That leaves the fill bar advancing with time, and time-based mirroring, unconfirmed.Not in scope
Distribution model, the React output, the other 13 themes, and any change to the shell's skin picker (extending
SKINSinapp/constants.tsripples into thesatisfies Record<\${SkinPreset}/${Skin}`, …>` maps and won't compile without package and registry entries).🤖 Generated with Claude Code