paper-grid
Skill75 starsChanged 40 days agoApache-2.0View on GitHub
Style for edit-video: a white paper pane with a quiet grid above the speaker and one clean animated UI artifact per spoken line (cards that type, counters, lists that expand, approvals that click). Use for tool, repo, skill or feature explainers where each sentence names something that can be drawn as UI.
- Installs packages
In this repository15 other files it gives its agents.
- AGENTS.mdAGENTS.md
- .claude/CLAUDE.mdCLAUDE.md
- find-ideasSkill
- write-scriptSkill
- edit-videoSkill
- animated-explainerSkill
- film-stylesSkill
- product-motion-adsSkill
- 3d-characterSkill
- cartoon-characterSkill
- adaptive-glassSkill
- breakout-cardSkill
- motion-effectsSkill
- text-over-footageSkill
- publish-and-analyzeSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 5 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 10
Every spoken item gets its own small, living artifact: a card types, a counter climbs, a row lights up, a cursor clicks.
- Distinctive: line 15
Keep the identity: white paper, the quiet grid, white cards with one rim, one navy accent, Helvetica Neue, one clear idea per beat, real …
- Distinctive: line 19
The library is a vocabulary, not a template: Use a component when it is the best picture of the line, and invent a new artifact when it …
- Distinctive: line 58
Keep it alive with typing, counters, a highlight sweep, a live dot.
- Distinctive: line 68
Don't reuse a reference creator's graphics or identity.
- Very specific: line 78
Write the run's build script: one with reel.scene(...) per beat (example_build.py shows the mechanics, not a storyboard):
- Very specific: line 86
Look before rendering: python3 scripts/snap.py <comp> <out> 0,40,120 for stills.
- Very specific: line 108
Put initial state in gsap.set.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Paper grid
A style on top of edit-video, which owns the cut, captions, sound and review. The
top half is white paper with a quiet grid; the speaker stays in the bottom half.
Every spoken item gets its own small, living artifact: a card types, a counter
climbs, a row lights up, a cursor clicks.
Fixed identity, open imagination
Keep the identity: white paper, the quiet grid, white cards with one rim, one navy accent, Helvetica Neue, one clear idea per beat, real proof for facts. Everything else is designed for the story in front of you.
- The library is a vocabulary, not a template. Use a component when it is the best picture of the line, and invent a new artifact when it isn't: a terminal that types, a diff that turns green, a flowchart that wires itself.
- Each reel should feel designed for its topic. Change which artifacts appear, how they connect, the rhythm and the one or two moments that surprise.
- Aim for one screenshot-worthy idea per reel: an object that transforms into the next beat, a satisfying click, a counter that lands on the spoken word.
Fits a talking head about a tool, repo, workflow or feature. Dark or saturated room footage benefits most: the white pane gives contrast and a clean seam. Not for personal reflections, jokes over footage or long screen demos that need the whole frame.
The look (starting defaults)
| Element | Setting |
| Paper | #fbfbfd pane at y 0–960. Grid 45px minor / 180px major lines at 5.5% / 10% ink, radial fade at the edges, drifting one cell over the reel |
| Seam | Hard split at y=960 with a 1px light line and a soft shadow into the camera |
| Cards | White, 30px radius, one 1px rim, soft drop shadow, about 740–820 wide, centered on x=540 |
| Chips | 64px white capsules with a 42px icon tile, 28px bold |
| Ink | #0f172a text, #64748b muted, navy accent #1e40af / #5b9eff. Red, amber and green only where they mean something |
| Type | Helvetica Neue. Card text 30–36px, labels 21–28px, hero figures 200–270px |
| Skeletons | Grey bars and gradient avatars stand in for people and body text; only the words that carry the point are real |
Tokens live in scripts/paper_grid.py (TOKENS), component CSS in
assets/paper-grid.css. Override per reel with Reel(tokens=...).
Layout and motion
- Split for almost everything. Artifacts sit at y 270–940; captions go on the
speaker's headroom (about y=.55). Set
split_video_topfrom the actual take so the head starts below the caption band. - Full camera only for re-hooks ("but the best part is…"): push in over the line, then return to the split on the reveal word.
- Frame 1 shows the product moving: open on the split with the first artifact already on screen and settling.
- One artifact per spoken item, on its word. It enters on the noun and acts on the verb. Entrances .35–.45s (cards rise 50px, chips pop), exits .2s, 6 frames before the next item. Carry one object through related beats.
- Keep it alive with typing, counters, a highlight sweep, a live dot; no perpetual shimmer, spin or bounce. Start from the motion library.
Truth rules
- Real proof for facts: the actual repo page, list or product screen
(
scripts/capture_page.py), checked against the source that day. Crop to the content that reads at phone size and push into the words that carry the beat. - Mockups stay generic: skeleton posts and counters carry no real names or
results; unmeasured meters carry an
ILLUSTRATIVElabel and no numbers. - Don't reuse a reference creator's graphics or identity. Borrow the mechanism only.
Build
Needs Python 3.10+ and the workspace runtime from edit-video's
init-workspace.mjs --install (HyperFrames and GSAP). Snapshots and page captures
also need pip install playwright and Google Chrome.
- Cut the dialogue with
edit-videoand keep a word timeline; beat frames come from it. - Capture proof into the run's
assets/, cropped to the fact that matters. - Write the run's build script: one
with reel.scene(...)per beat (example_build.py shows the mechanics, not a storyboard):sys.path.insert(0, '<this skill>/scripts') from paper_grid import Reel reel = Reel('<run>/work/grid-v01', camera='<run>/assets/camera.mp4', total=1314)compile()writes the HyperFrames folder and checks the timeline script. - Look before rendering:
python3 scripts/snap.py <comp> <out> 0,40,120for stills. - Render with
.favstash-studio/runtime/node_modules/.bin/hyperframes render <comp> --fps 30 --quality high --output <raw>.mp4, thenscripts/fix-frame0.sh <raw> <visual>.mp4if frame 0 came out black. - Assemble dialogue, captions and sound in your editor or with FFmpeg, and review
with
edit-video.
Sound map (edit-video's SFX pack)
| Visible event | Cue |
| Frame 1 | hook-impact |
| Split ↔ full camera | camera-shutter; the light shutter for a smaller change |
| Re-hook build | metallic-riser, resolving on the reveal |
| Card, chip or pill arrival | ui-pop (merge arrivals under ~6 frames apart) |
| Toggle, highlight, button press | ui-click |
| Expand, shrink, fly-away | zoom-swish |
| Typed text | typing-tick.wav, one cue per ~0.9s of typing |
| Earned approval | clean-ding.wav, once |
Pitfalls already paid for
- Anything that enters later must start hidden (
opacity:0), or it shows before its cue. - Put initial state in
gsap.set; atl.setat 0 alone can revert on seek. - A
<video>given only a width renders at the file's pixel height and letterboxes. Set an explicit height withobject-fit:cover. - Give every component a distinct id; the library's own layers use
root,top,grid,seam,camFull,camSplitand theirVvideos. - Snapshots can race on seek. Trust the encoded render when they disagree.
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.

