duo
Skill155 starsChanged 3 months agoApache-2.0View on GitHub
Composite any two HTML screens into a locked photo plate of hands holding an open foldable phone — a 1448×1086 still with two blank 495×849 / 498×849 screen slots (left and right) — to make side-by-side meme and comparison videos: TikTok vs Reels, ChatGPT vs Claude, "me vs my friend", before/after, two apps racing. Trigger on: "foldable phone meme", "put X on the left screen and Y on the right", "two-screen comparison video", "the dual-screen phone meme", or any request that names two apps/feeds/chats to show at once. The plate, geometry, and camera are finished — the agent only authors what plays INSIDE the two screens and renders.
4 other projects use a version of this.
Adopted by
Most recently taken up by
- yamancan/skillstheir own version, first seen Oct 20, 2026
- Audatic07/duotheir own version, first seen Oct 2, 2026
- agentik-os/claude-code-skillstheir own version, first seen Sep 17, 2026
- juanexpensive/Skillstheir own version, first seen May 25, 2026
Other versions4 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 4
- Audatic07/duo★ 22same name
- yamancan/skills★ 2same name
- agentik-os/claude-code-skills★ 1same name
- juanexpensive/Skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository7 other files it gives its agents.
- camera-3d-captionsSkill
- day-in-my-lifeSkill
- p5-paint-animationSkill
- prod-by-claudeSkill
- session-storySkill
- vox-explainerSkill
- x-posting-licenseSkill
The file
We split this skill into the rules it gives an agent
6 rules. 0 are common, 5 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
5
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 37
Use for any "two things side by side on a phone" beat: app vs app, feed vs feed, two chats answering the same prompt, two versions of a …
- Distinctive: line 39
Do not use when the user wants a single phone, a laptop, or a real device bezel they supply — this skill is one specific photo.
- Very specific: line 62
No credentials, no paid operations, no telemetry.
- Distinctive: line 111
Render: only when the user asks (ask first if your harness gates renders):
- Distinctive: line 126
Never resize, move, or re-radius the screens to "fit" content.
- Distinctive: line 133
Keep the camera push as the only rig motion.
“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Duo
One fixed 1448×1086 @ 30fps composition: a photo of two hands holding an open foldable, with both screens blanked. You fill the two screen slots with HTML (live video, a mock chat, a scrolling feed, anything HyperFrames renders) and the rig masks it to the phone's true rounded-corner geometry.
Image: plate
The phone and hands are a photograph, not something you build.
assets/plate.png IS the device. Never recreate, redraw, or CSS-model the
foldable, its bezel, the hinge, or the hands, and never substitute a different
image. build.mjs copies the plate into every project; if it is missing, stop
and reinstall the skill rather than improvising a phone.
Do not edit the rig. The two .screen rects, their corner radii and the
hinge divider were measured from that photo pixel by pixel. Every visual change
you make happens inside #screen-left .content and #screen-right .content,
plus the one shared GSAP timeline.
When to use / when not to
- Use for any "two things side by side on a phone" beat: app vs app, feed vs feed, two chats answering the same prompt, two versions of a product.
- Do not use when the user wants a single phone, a laptop, or a real device
bezel they supply — this skill is one specific photo. For a different device
plate you would re-measure geometry the same way (see
references/screen-geometry.md) rather than stretch this one.
Requirements
- Node 18+ and the HyperFrames CLI via
npx hyperframes@latest(@latestis mutable; pin a version for byte-identical re-renders). - Load your motion doctrine first if your workspace has one. The rig's only built-in motion is one slow camera push; everything inside the screens must perform (scrolls, swipes, typing, taps), never idle-wobble.
Network and side effects (complete list)
registry.npmjs.org— the HyperFrames CLI itself, vianpx.cdn.jsdelivr.net— the composition loads GSAP (pinned3.14.2) at preview/render time. Rendering is not fully offline.- Anything YOU add to the screens (video clips, avatars, fonts) is your side
effect: fetching a site, downloading media, or scraping UI happens only if the
user's request calls for real content, and only from sources they are entitled
to use.
references/sourcing-real-ui.mddocuments how, and the rights caveat.
No credentials, no paid operations, no telemetry. build.mjs writes only
inside --out and refuses to write through symlinks.
Flow
<SKILL_DIR> is this skill's installed directory (e.g.
~/.claude/skills/duo).
- Scaffold a project:
node <SKILL_DIR>/scripts/build.mjs --out ./foldable-meme --duration 10 --icons tiktok,instagram--icons is optional; it copies the real TikTok / Instagram glyph sheets
(assets/icons-*.svg, extracted from the live mobile sites) into the project.
- Decide what each screen does for the length of the clip, in writing, before touching HTML. The pause test: at any second, something inside at least one screen must be mid-motion (a swipe landing, a video playing, text arriving). Two static screenshots is not a video.
- Author the screens. Inside each
.contentyou have a normal DOM viewport: left495×849, right498×849. Position against those, never the root. Patterns (full-screen snap feed, drag→fling swipe, like-tap, chat typing) are inreferences/feed-recipes.md. Rules:<video class="clip" data-start data-duration muted playsinline>for every clip; HyperFrames owns playback. Overlap the windows of an outgoing and incoming clip across a swipe so both are decoded mid-transition.- One timeline. Add your tweens to
tlinside the marked hook; do not create a secondgsap.timelineor register anotherwindow.__timelineskey. - No
repeat: -1, noMath.random, no clocks — the render seeks. - Status bar: the photo's original screens showed
9:41top-left of the left screen and wifi/battery top-right of the right screen. Reproduce that split if your apps show a status bar; it sells the "one device" read.
- Check and snapshot:
npx hyperframes@latest check ./foldable-meme
npx hyperframes@latest snapshot ./foldable-meme --at 1,4,7 --no-endLint must be 0 errors. Layout/contrast warnings about text passing under a fixed header, or white UI text over video, are expected for app UIs; read the frames instead and confirm (a) content reaches every rounded corner with no gap, (b) the gray hinge divider is visible between the screens, (c) nothing from one screen leaks over the hinge.
- Render only when the user asks (ask first if your harness gates renders):
npx hyperframes@latest render ./foldable-meme -o ./foldable-meme/renders/v1.mp4Verify success
check passes lint with 0 errors; snapshot frames show both screens filled
edge to edge inside the phone with the divider intact; the MP4 is the requested
duration at 1448×1086 and, on a pause at any second, something on screen is
moving. Pull frames from the rendered MP4 (not the preview) for the final QC.
Rules
- Never resize, move, or re-radius the screens to "fit" content. Fit content to the slots. If a mock app looks too big, scale the app's internal type, not the screen.
- Real content must be real: if the user asks for TikTok or Instagram, pull the actual UI glyphs and layout metrics from the live site and use real posts (with their real usernames and counts), not invented ones. Invented handles are only for fictional / parody screens the user explicitly wants.
- Keep the camera push as the only rig motion. Do not add drift, breathe, or glow loops anywhere; fill time with story inside the screens instead.
- Output stays 1448×1086 (the plate's native 4:3). Offer a 16:9 or 1:1 crop as a follow-up render rather than resizing the rig.
Files
assets/plate.png— the clean plate, screens blanked to neutral (1448×1086).assets/screen-spec.json— measured rects, radii, hinge, status-bar split.assets/icons-tiktok.svg,assets/icons-instagram.svg—<symbol>sheets of the real mobile-web glyphs (like, comment, share, nav bars…).template/index.template.html— the locked rig with two empty slots.scripts/build.mjs— scaffolder.references/screen-geometry.md— how the geometry was measured and how to re-measure for a new plate.references/feed-recipes.md— snap feed, swipe physics, like tap, chat typing.references/sourcing-real-ui.md— pulling real app UI and clips, and the rights caveat.
Provenance and licensing
- The plate is derived from a photo posted publicly on X by @A_Kapustin
(status
2097769962936868872, a foldable showing ChatGPT and Claude side by side); the two screens were blanked so it can carry any content. It is redistributed here as a meme template; users of this skill publish results at their own judgment and should credit the source where appropriate. - The icon sheets reproduce TikTok's and Instagram's own UI glyphs (extracted from their public mobile web DOM) so mock screens look like the real apps. They are trademarks of their owners and are included for parody / commentary use only.
- GSAP is loaded from jsDelivr under its standard license; it is not vendored.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

