agentleFS
ScrapybaraScrapybara/agent-skills/skills/banger-video/SKILL.md

banger-video

Skill0 starsChanged 49 days agoMITView on GitHub

Makes premium product launch and feature videos (15–25 s, 1080p60, cut on the beat) as HTML + GSAP compositions rendered with HyperFrames. Use when asked for a launch video, feature video, promo, teaser, product montage, sizzle, or "a video like <some launch clip>" for any product.

  • In this repositoryOne other file it gives its agents.

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 7 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Product display, never a demo: No mouse cursor driving the UI, no clicking through settings, no screen-recorded walkthroughs, no narration …

    line 24
  2. Distinctive:

    No raw screenshots in the frame: No arbitrary rectangle crops of screenshots, no spotlight / "dim the rest" highlights, no glow rings, and …

    line 27
  3. Distinctive:

    One concept per video: Every video in a set gets its own idea and look, such as a light flat diagram, a dark living document, or a …

    line 34
  4. Distinctive:

    Never a paragraph, an expanded comment card, or a diff the viewer has to parse.

    line 39
  5. Very specific:

    Research first: Before designing, find launch clips of the same kind of feature from the best companies and study them frame by frame …

    line 41
  6. Distinctive:

    Never cut mid-animation: Before cutting any footage, including your own renders, map it into transitions and settled states at 30 fps.

    line 44
  7. Distinctive:

    Stay on topic: If the story is "make X", the payoff shows only X.

    line 49
  8. Distinctive:

    Verify every frame yourself: before delivering: contact sheets at 4–10 fps over the whole cut, 30 fps strips around every cut and hero …

    line 58
  9. Very specific:

    Real content: use the product's real strings (names, IDs, file names, statuses), design tokens, fonts and icons.

    line 68
  10. Very specific:

    Check: run npx --yes hyperframes lint, then snapshot --at …, look at the frames, and iterate.

    line 76
  11. Very specific:

    Render the picture: run render -q high -f 60 -w 2 as a background job.

    line 78

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Banger video

Short product videos in the style of the best software launches: a product display, not a demo. They are built as HTML and GSAP, rendered to MP4 with the HyperFrames CLI, and cut on the music's beat grid. The bar is whether a client would pay $10k for it, so work as a motion designer and editor, not a screen recorder.

Rules

Each rule came from a rejected cut.

  1. Product display, never a demo. No mouse cursor driving the UI, no clicking through settings, no screen-recorded walkthroughs, no narration captions. If the product has an agent that acts, show the agent acting, not a person operating the app.
  2. No raw screenshots in the frame. No arbitrary rectangle crops of screenshots, no spotlight / "dim the rest" highlights, no glow rings, and no drifting 3D screenshot planes with fake depth of field. Rebuild the UI as clean HTML graphics from the product's own design tokens. A real capture is allowed only as a whole component grabbed from the actual element (references/capture.md). Mock apps that look fake are worse than no apps.
  3. Pixel-exact. Anything that marks UI (tags, outlines, avatars, badges) is positioned from measured geometry of the laid-out page, never eyeballed.
  4. One concept per video. Every video in a set gets its own idea and look, such as a light flat diagram, a dark living document, or a full-frame montage. Never reuse one template across features.
  5. Show the product doing the work, with nothing to read. The payoff is a visible change of state: lines swap red→green, tags flip to "✓ Resolved", statuses turn green, a counter completes. Never a paragraph, an expanded comment card, or a diff the viewer has to parse. Use one short line per beat at most.
  6. Research first. Before designing, find launch clips of the same kind of feature from the best companies and study them frame by frame (references/research.md). Name the clip whose structure you are borrowing.
  7. Never cut mid-animation. Before cutting any footage, including your own renders, map it into transitions and settled states at 30 fps. A cut enters on a settled frame or the exact start of a transition, and exits on a settled frame. A hero change, such as several items resolving, is one continuous shot: speed-ramp the dull parts 1.6–2× and keep each state change at real speed, landing on a beat or half-beat.
  8. Stay on topic. If the story is "make X", the payoff shows only X. Don't pad it with shots from other videos.
  9. Montages: the first and last item hold for about 1.5–2 s, and everything between is split-second (one beat or half a beat). Keep them full frame, with no zoom or push-ins, no per-item labels, and no end-card slogan unless asked.
  10. Music leads, sound design supports, and you choose. Pick the track yourself with the method in references/music-and-sound.md rather than sending rounds of previews. Pop, hyperpop, anime songs and generic corporate beds all read wrong for a launch, and so does a quiet bed buried under sound effects.
  11. Verify every frame yourself before delivering: contact sheets at 4–10 fps over the whole cut, 30 fps strips around every cut and hero moment, and full-res stills of key frames. If you can no longer look at images, hand the check to a fresh agent rather than ship unchecked.

Workflow

  1. Brief in one message: the feature, a one-sentence story of what the product does, the reference clip, the look, and the music.
  2. Research (references/research.md): pull 5–15 relevant launch clips and write down what the best ones do.
  3. Real content: use the product's real strings (names, IDs, file names, statuses), design tokens, fonts and icons. If it's a web app you can run, capture exact geometry and whole components from it (references/capture.md).
  4. Music: choose and time the track (scripts/music.py, references/music-and-sound.md). Fix the lift offset and the beat period p before writing any timeline.
  5. Build (references/hyperframes.md): a build.py holds the beat schedule (B(n) = D + n*p), generates index.html from a template, and writes an events.json of every moment that should make a sound. Measure the real layout into widths.json once, then rebuild.
  6. Check: run npx --yes hyperframes lint, then snapshot --at …, look at the frames, and iterate.
  7. Render the picture: run render -q high -f 60 -w 2 as a background job. More workers can exhaust a 16 GB machine.
  8. Mix and mux the audio yourself (scripts/mix.py, then ffmpeg -c copy). The renderer's audio path can change levels.
  9. Deliver: attach the MP4 and state its duration, the track, and its licence terms (attribution text, paid-ads limits).

Beat sheets that worked are in references/patterns.md.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.