promo-video
Skill2 starsChanged 48 days agoMITView on GitHub
Makes promotional, launch and explainer videos as code (motion graphics, kinetic type, real product screen footage, 3D with three.js, fluid ink and a composed or supplied soundtrack), rendered frame-exact to MP4 for LinkedIn, X, YouTube, Instagram, talks or a website. Starts with a short interview, storyboards stills before anything moves, and ships a review page for comments. Use when the user wants any promo video, teaser, showreel, launch or announcement video, product demo video, animated explainer, or a video for a paper, tool, app or release, or wants to re-cut, restyle or add music to a video made this way. Triggers on "make a video", "a small video for my LinkedIn post", "something to post", "a teaser for the release". Not for trimming or converting an existing recording (a plain ffmpeg job) or for still images and slide decks.
- Pipes a download into a shell
- Installs packages
19 other projects use a version of this.
Adopted by
Most recently taken up by
- CatCatUncle/openworkbuddytheir own version, first seen Nov 5, 2026
- chessunyan/renoise-skillstheir own version, first seen Nov 1, 2026
- thesleax/video-kittheir own version, first seen Nov 1, 2026
- qxryz/workflowgeneratortheir own version, first seen Oct 25, 2026
- prajwal-svm/product-promo-harnesstheir own version, first seen Oct 20, 2026
- ryokun6/ryostheir own version, first seen Oct 17, 2026
Other versions19 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 19
- ryokun6/ryos★ 1.3ksame name
- pedronauck/skills★ 655same name
- CatCatUncle/openworkbuddy★ 413same name
- OpusGameLabs/game-creator★ 360same name
- AKCodez/promo-video-skill★ 53same name
- qxryz/workflowgenerator★ 48same name
- iart-ai/ecommerce-video-skills★ 17same name
- smit-vanani/claude-promo-video★ 14same name
- freema/ai-skills★ 5same name
- divakarwl12/skills★ 4same name
9 more not listed
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 repository23 other files it gives its agents.
- bounded-goalsSkill
- change-asserted-valueSkill
- check-targetSkill
- cluster-failuresSkill
- councilSkill
- enumerate-firstSkill
- eval-loopSkill
- fable-reviewSkill
- gauntlet-loopSkill
- handoffSkill
- kickoffSkill
- prove-invariantSkill
- refusal-censusSkill
- reuse-before-rerunSkill
- run-evalSkill
- run-gateSkill
- scientific-posterSkill
- spy-siteSkill
- test-gateSkill
- trace-frictionSkill
- understand-before-mergeSkill
- verify-sourceSkill
- watching-background-jobsSkill
The file
We split this skill into the rules it gives an agent
12 rules. 0 are common, 5 are distinctive, and 7 are very specific to this project.
0
Common
Most of the 16 edited versions we read keep these.
5
Distinctive
This author's own choices. Worth a look.
7
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 20
Default to Level 2. Go straight to Level 1 only when the user asks for it ("just make it", "surprise me", "one-shot", a showreel).
- Very specific: line 36
Before asking anything, learn what you can: read the project's README, website, design tokens (CSS custom properties, a design system …
- Distinctive: line 53
Ask for them; never invent them.
- Very specific: line 62
Colour scheme: always ask, with a small preview per option (the picker's preview field): the brand's own palette first, then 2-3 real …
- Distinctive: line 64
Treat any colour the user rules out as a hard constraint everywhere you draw, 3D materials and fluid ink included.
- Distinctive: line 66
Fonts: always ask, previewing each pairing (headline / text / label): the brand's own fonts first, then 2-3 alternatives with a different …
- Very specific: line 69
Write the answers into brief.md in the job folder: message, audience, format, length, facts (verbatim, with where each came from), CTA, …
- Very specific: line 89
If setup.sh prints MISSING, install that tool and re-run Step 1.
- Very specific: line 99
Write a shot list in brief.md: each shot's start time, duration, caption, what moves, sound.
- Very specific: line 163
When the user likes a piece (a logo reveal, a transition, a window treatment), save it to ~/Movies/promo-videos/_library/<name>/ as a small …
- Very specific: line 171
Blank 3D or fluid canvas: read the PAGE ERRORS render.mjs prints.
- Distinctive: line 176
Honesty: only real UI, real facts, real partners.
“Common” is measured: kept by at least 6 in 10 of the 16 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
Promo video
A video here is a web page: HTML, CSS and JavaScript on a fixed-size stage, where every frame is
a pure function of time. scripts/render.mjs calls renderFrame(t) frame by frame, screenshots
each one into ffmpeg, and muxes a soundtrack the stage composed in code. That makes the video
deterministic, editable like code, and as sharp as the browser can draw.
The work runs in three levels:
- One-shot: build the whole video from one brief. Fast; good for showreels and tests.
- Storyboard (the default): stills of every shot with timings, which the user comments on before anything moves. It catches the wrong video a third of the way in instead of at the end.
- Directing: the user reviews the real video on a review page, pinning comments to moments (picture and sound) and moving cuts, then you apply exactly those changes.
Default to Level 2. Go straight to Level 1 only when the user asks for it ("just make it",
"surprise me", "one-shot", a showreel); Level 1 skips Step 2. Re-cutting a job made here (new
music, restyle, trims)? Read its brief.md and stage.html, skip Steps 0-1, and start at the
step the change needs.
Copy this checklist and track your progress:
- 0 Research done,
brief.mdwritten and shown - 1 Job folder ready, setup.sh printed no MISSING
- 2 Storyboard approved (Level 2)
- 3 Footage captured (live product only)
- 4 Self-check clean: frames looked at, about -14 LUFS, no PAGE ERRORS
- 5 Review notes applied, user says the version is done
- 6 Final MP4 and cover delivered
Step 0 - Research, then interview
Before asking anything, learn what you can: read the project's README, website, design tokens
(CSS custom properties, a design system file), and find logos and brand assets (public/,
assets/, docs/, logos/). If a live site exists, look at it. Questions are for what cannot be
looked up, and they are cheaper when you can offer the right options.
Then interview with the question picker (AskUserQuestion), 2-4 questions per round, each with a
recommended option first and enough context to pick fast. Keep asking in rounds until every item
below has a clear answer: an answer that raises a new question (a free-text note, a constraint
such as "no blue", a choice that conflicts with the brand or with chemistry conventions) gets
its own follow-up round before you build anything. Skip any question the conversation or the
research already answered, and never ask for what you could look up. Cover:
- Goal and message: what the video must make a viewer think or do; the one-line promise.
- Where it runs: platform and format (see
references/motion-craft.md; default LinkedIn 4:5). - Length: 15 s teaser / 30-45 s feature / 60 s walk-through.
- Audience: peers in the field, customers, general public, funders.
- Facts: the real features, numbers and claims to show, and the call to action (URL, button). Ask for them; never invent them. Offer the ones you found and let the user correct.
- Brand: confirm the logos, colours and fonts you found; partner logos only if stated.
- Footage: is there a live app or site to film, and which flow shows it best?
- Sound: for anything that will be posted, recommend a real track (the user generates one in
Suno or picks a licensed one, from a prompt you write) and fit the cuts to it
(
references/music.md, "Fitting a supplied track"). Code-made music with a mood is the fallback and fine for drafts; be upfront that it sounds synthetic. - Look and energy: calm and premium vs bold and fast; 3D and fluids used where they carry the story (default), everywhere, or not at all.
- Colour scheme: always ask, with a small preview per option (the picker's
previewfield): the brand's own palette first, then 2-3 real alternatives (dark vs light ground, accent sets). Treat any colour the user rules out as a hard constraint everywhere you draw, 3D materials and fluid ink included; footage of a real product keeps its real colours, so say so. - Fonts: always ask, previewing each pairing (headline / text / label): the brand's own fonts first, then 2-3 alternatives with a different character (condensed, heavy sans, grotesk, serif).
Write the answers into brief.md in the job folder: message, audience, format, length, facts
(verbatim, with where each came from), CTA, brand, footage plan, sound, look. Show it in one
short summary and continue unless the user objects.
Step 1 - Job folder and tools
Needs Node.js (current LTS, with npm: https://nodejs.org) and ffmpeg (brew install ffmpeg or
sudo apt install ffmpeg); uv (curl -LsSf https://astral.sh/uv/install.sh | sh) runs the
librosa and RDKit snippets. setup.sh fetches playwright-core and Chrome itself. On Linux, use
xdg-open where this file says open.
SK=<this skill's base directory> # e.g. ~/.claude/skills/promo-video
JOB=~/Movies/promo-videos/<slug> # outside any git repo: frames and renders are large
mkdir -p "$JOB" && cd "$JOB"
for f in stage-kit.js fluid.js score.js; do [ -e "$f" ] || cp "$SK/assets/$f" .; done
[ -e stage.html ] || cp "$SK/assets/stage-template.html" stage.html # never overwrite a job's stage
bash $SK/scripts/setup.sh # finds or fetches Chrome, playwright-core, ffmpegIf setup.sh prints MISSING, install that tool and re-run Step 1. Execute the scripts in
scripts/ (lib.mjs is their shared helper, executed by the scripts). Read the header comment of
each copied .js file for its API, and do not edit them. In stage.html, keep the render
contract in its header comment exactly and replace the three placeholder shots.
Copy the project's logos and assets into $JOB/assets/. Keep everything the stage loads next to it;
the stage is opened from file://.
Step 2 - Storyboard (Level 2)
- Write a shot list in
brief.md: each shot's start time, duration, caption, what moves, sound. Plan it in bars of the chosen tempo so cuts can land on beats (references/music.md). - Build
stage.htmlin the final look: layout, type, colours, logos, 3D objects placed, but motion can be rough. Readreferences/motion-craft.mdfirst; readreferences/3d-and-fluids.mdif a shot uses three.js or fluid ink. - Render one still per shot at the moment that best shows it:
node $SK/scripts/render.mjs stage.html --stills 0.6,4.2,8.0 --outdir board - Copy
$SK/assets/storyboard.htmlinto the job and writeboard-data.js:window.BOARD = { title, format, duration, shots: [{ id, t, img, title, desc, sound }] }. open storyboard.htmland stop. Tell the user: click a still to pin a comment, then Copy all comments and paste them back.
Comments come back as lines like Shot 07, 5.90s, pin at 52% across, 46% down: less text heavy.
Apply every comment and change nothing else, re-render the affected stills, and ask for another
pass or approval. Approval is the user saying so, or a pasted "No comments: the storyboard is approved."
Step 3 - Footage (when there is a live product)
Copy $SK/assets/shots-template.mjs to shots.mjs, script the flow, and run
node $SK/scripts/capture.mjs shots.mjs --outdir footage (a new --outdir per take: capture.mjs
empties its frames folder). Each mark(key, selectors) records a marker the stage hangs a caption
and a camera zoom on. In the stage:
const F = Kit.footage(window.DATA.footage, [['start', 1, null], ['result', 1.6, null], ...]),
then await F.apply(img, t - APP0, windowWidth) inside renderFrame, and render with
--data footage=footage/footage.json. Present footage in a floating browser window with captions
in a band above it (motion-craft.md, "Real product footage").
Step 4 - Build the motion and the sound
- Animate with
Kit(tween,ease,riseLines,timeline,stepper): everything fromt. - Compose with
Scorein the stage and setwindow.renderAudio = () => S.render()(references/music.md). For a user track, leaverenderAudiounset and render with--track music/<file>. - Preview sections quickly while working:
node $SK/scripts/render.mjs stage.html --out part.mp4 --from 8 --to 16 --fps 15. - Self-check before showing anything (motion-craft.md, "Self-check"): contact sheet, full-size frames at each shot, spectrogram, loudness, no PAGE ERRORS (render.mjs then exits 2). Look at the images; do not assume. If any item fails, fix the stage, re-render that section and repeat the self-check before Step 5.
Renders are fast (a 60 s video with 3D is typically 1-3 minutes). Run anything longer than about two minutes in the background and keep working or report status.
Step 5 - Review (Level 3)
- Full render:
node $SK/scripts/render.mjs stage.html --out review.mp4(add--data,--track). - Copy
$SK/assets/review.htmlinto the job, writereview-data.js:window.REVIEW = { title, video: 'review.mp4', fps: 30, scenes: [{ id, title, start, end }] }. open review.html. The user plays it at 1x or 2x, clicks the picture to pin notes, adds sound notes, moves shot starts and ends to the playhead, and pastes back Copy all.- Apply every note and trim, change nothing else, re-render, and let them review again. Do not make the final render until the user says the version is done.
Step 6 - Final delivery
- Render the final:
node $SK/scripts/render.mjs stage.html --out <slug>.mp4(add--ss 2for crisper thin lines). - Save a cover image at the frame that sells the video best (the platform lets the user pick it):
ffmpeg -ss <t> -i <slug>.mp4 -frames:v 1 <slug>-cover.png. - Copy both to the Desktop (or where the user wants them) and report: length, size, format, where.
- Offer post text in the user's voice, built only from the brief's facts.
Reusable elements
When the user likes a piece (a logo reveal, a transition, a window treatment), save it to
~/Movies/promo-videos/_library/<name>/ as a small stage snippet with a note on how to reuse it,
and check the library at the start of the next job.
Things that bite
- macOS privacy: files in
~/Downloads,~/Desktopor~/Documentsmay be unreadable ("Operation not permitted"). Ask the user to move the file into the job folder. - Blank 3D or fluid canvas: read the PAGE ERRORS render.mjs prints; see 3d-and-fluids.md, "Headless GPU".
- Fonts: load them with
display=block; render.mjs waits fordocument.fonts.ready. - A site that looks different on real devices (WebGPU, touch, 120 Hz): the capture is a desktop Chrome; say so if the promo claims tablet or phone behaviour.
- Honesty: only real UI, real facts, real partners. A promo that overstates is a liability for the person who posts it.
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.

