toolkit-video
Skill9 starsChanged 13 days agoMITView on GitHub
Build a fast montage video that shows off a pack of slash commands or skills, a clean white prompt bar types each command, the autocomplete menu shows its one-line description, Enter, and the result plays and fills the screen, one command every 2 seconds, ending on a grid of every command and a comment keyword. For a feature list, a service menu, a resource roundup, a skill pack. Use when someone says "slash command video", "show my skills in a video", "skill pack promo", "prompt bar video", "show what each command makes" or "giveaway video for my pack".
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
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 33
Music only: No voice needed. Every word on screen must read with the sound off.
- Distinctive: line 42
Use a product's real logo only if the video is about that product.
- Very specific: line 66
Run video-check. Set up the keyword delivery before you post (see call-to-action).
≈ “Common” is an estimate: at least 1 in 5 of the 308,232 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
Toolkit video
Use it for: a feature list, a service menu, a resource roundup, a skill pack.
Show a pack of commands the fastest way possible: type it, see what it makes, next. No talking needed.
Before you start: each command needs a real result to show (a clip or a still). Collect them first. A command with nothing to show stays out of the montage and goes on the end grid only.
Every size. The same build makes 9:16, 4:5, 1:1 and 16:9. Size objects from PANEL, then run every-format.
Step 1: Ask for four things, then wait
- The commands, in order, each with a one-line description (under 50 characters).
- One real result per command: a 2-3 second clip or a still. Your own work only.
- The comment keyword, 4 or 5 letters, and what it sends.
- The text hook, two lines (see
opening-hook).
The format
| Time | On screen |
| 0-1.8s | Text hook, big, two lines. The prompt bar sits under it, cursor blinking |
| Every 2.2s | / typed, the menu opens with the command and its description highlighted, Enter, the result plays inside the bubble, then fills the frame for about 1s |
| Last 4s | The keyword typed into the bar. A grid of every command zooms out. "Comment KEYWORD" holds to the last frame |
- 8 to 9 commands in the montage keeps it near 25 seconds. The rest go on the end grid.
- Lead with the commands people want most. The strongest result goes second, right after the hook.
- Music only. No voice needed. Every word on screen must read with the sound off.
The prompt bar (the restyle)
A clean white composer, not a dark phone app:
- White page, the bar centred at y 1320, 960px wide, 36px corners, hairline border, soft long shadow.
- Placeholder text in grey. Typed text
#1d1d1f, 46px, monospace for the command. - The menu rises above the bar: white card, 28px corners, each row a command in monospace and its description in grey. The chosen row has a pale accent fill.
- The send button is a filled circle in your accent colour (
#0071e3default). It presses down on Enter. - Use a product's real logo only if the video is about that product. Never a lookalike icon.
Step 2: Build brief (paste into Claude Code)
Build one HTML file, index.html, 1080x1920, with window.seek(seconds) that draws any frame from time alone, and ?render to wait for seek.
White #ffffff page, text #1d1d1f, accent #0071e3, SF Pro or Inter.
0-1.8s: my two-line hook at 96px, centred at y 520. Under it, the prompt bar (y 1320, 960px wide, 36px corners, hairline, soft long shadow) with a blinking cursor.
Then for each command in my list, a 2.2s cycle:
0.0-0.5s type "/" then the command at 16 characters a second, monospace 46px.
0.2s after "/", a menu card rises above the bar showing 4 rows: the command (highlighted with a pale #0071e3 fill) and 3 neighbours, each with its grey description.
0.6s Enter: the send button presses, the menu closes, a result bubble grows above the bar playing my clip for that command.
1.2-2.2s the bubble scales up to fill the frame (snap in, settle), then shrinks back as the next "/" is typed.
Last 4s: type "[KEYWORD]", then a grid of every command card zooms out, and "Comment [KEYWORD]" sits in a white pill at y 962 to the last frame.
Results come only from my files: clips/<command>.mp4 or .jpg. Never draw a fake result.
Check stills at the moment each result fills the frame and at the last frame.For clips inside a seek page, pull frames first (ffmpeg -i clip.mp4 -vf fps=30 -q:v 3 clips/<command>/%06d.jpg) and pick the frame from time, so the render is frame exact.
Step 3: Check
- Can someone read every command and its description in the 2.2s it is up?
- Is every result real and made with that command?
- Run
video-check. Set up the keyword delivery before you post (seecall-to-action).
What I learned the hard way
- The comment gate drives this format. On the best reels in this style, comments outnumber likes.
- Each result needs under 2 seconds to make sense. Pick the moment that shows the payoff, not the build-up.
In this repository40 other files it gives its agents.
- animated-numbersSkill
- apple-styleSkill
- behind-my-headSkill
- big-idea-videoSkill
- brand-setupSkill
- call-clipsSkill
- call-to-actionSkill
- clone-roomSkill
- cube-spinSkill
- depth-uiSkill
- every-formatSkill
- explainer-videoSkill
- frame-breakSkill
- grid-diveSkill
- head-pop-outSkill
- infinite-zoomSkill
- liquid-glassSkill
- magnifying-lensSkill
- news-reactionSkill
- opening-hookSkill
- phone-fanSkill
- portalSkill
- product-demoSkill
- pull-back-revealSkill
- ribbon-wrapSkill
- screen-in-screenSkill
- shatterSkill
- spotlightSkill
- stripe-styleSkill
- tier-listSkill
- tile-tunnelSkill
- title-orbitSkill
- vertigoSkill
- video-checkSkill
- video-effectsSkill
- video-intakeSkill
- video-soundSkill
- word-highlightSkill
- zoom-cutSkill
- zoom-throughSkill
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
Reviews can't be read right now.
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.

