hoodini/ai-agents-skills

video-to-landing-page

Turn any video into a cinematic scroll-driven landing page — Apple-style hero where scrolling progresses the visible frame through the video.

Quelltext ansehen

Verwendung

  1. Installationsbefehl kopieren und das Quellpaket vor der Installation in der eigenen Agentenumgebung prüfen.
  2. Benötigte Werkzeuge, Zugangsdaten und Hosts im Original prüfen.
  3. Eingabe und gewünschtes Ergebnis beschreiben und die Ausgabe mit den Quellen abgleichen.
npx skills add https://github.com/hoodini/ai-agents-skills --skill video-to-landing-page

Vor dem Start

  • Die Installation umfasst weder Modellabo noch API-Kontingent oder Dateizugriffsrechte.
  • Sterne und Installationen zeigen Aktivität, keine unabhängige Qualitäts- oder Sicherheitsprüfung.
Originales Skill-Dokument

Aus dem Quell-Repository gerendert; Überschriften, Beispiele, Code, Tabellen, Links und Bilder bleiben erhalten.

Video → Scroll-Driven Landing Page

Build a cinematic landing page where the hero is a video playing on scroll instead of time — like Apple product pages or the GitHub Copilot landing. Frames are extracted from the source video and swapped as the user scrolls; below the sticky hero, normal-scroll sections carry copy and CTA.

When to invoke

  • User drops a video and asks for a landing page, scrolling site, scroll-frame hero, or website from a video
  • Mentions of "Apple-style scroll", "GitHub Copilot landing", "scrub-on-scroll", "frame on scroll"
  • Hebrew equivalents: "דף נחיתה מסרטון", "אתר עם גלילה", "סרטון בגלילה"

Save location

Default: ~/Documents/yuv-projects/landings/<slug>/ — always save landing pages here so you can find them again. Create the parent directory if missing.

bash
mkdir -p ~/Documents/yuv-projects/landings
cd ~/Documents/yuv-projects/landings
# Build <slug>/ here.

Final path: ~/Documents/yuv-projects/landings/<slug>/. Tell the user where the landing lives at the end of every build.


Workflow (8 steps)

  1. Probe the source — ffprobe for duration, fps, dimensions, audio.
  2. Pick frame count — default 80. Floor 40 (choppy if lower), ceiling 160 (heavy page). Choose by duration: ≤10s → 50, 10–30s → 80, 30–60s → 120, >60s → 160.
  3. Decide output directory — default ~/Documents/yuv-projects/landings/<slug>/. The <slug> is short, derived from the video name or the topic. Override only on explicit request.
  4. Extract frames — run references/extract-frames.py <source-video> <output-dir> [count]. Outputs output-dir/frames/f_0001.jpg… with consistent JPG quality 82 and a max-width of 1920 (downscaled if source is larger).
  5. Copy and customise references/landing-template.html into output-dir/index.html:
  • Replace the __FRAMES__ placeholder with <img> tags for each frame (the build script does this for you when you pass --build-html).
  • Fill in __HEADLINE__, __TAGLINE__, __CTA_TEXT__, __CTA_HREF__, sections.
  • For Hebrew content: set <html lang="he" dir="rtl"> and swap fonts to Rubik/Assistant (see references/design-notes.md).
  1. Optimise frames — the build script already scales to 1920px max-width. If the page total weight is > 12 MB, drop quality to 75 or reduce frame count.
  2. Preview — python -m http.server inside the output directory and open http://localhost:8000. Scroll feels buttery on a decent machine.
  3. Deploy — drop the folder on Vercel/Netlify/Cloudflare Pages. No build step needed.

The scroll mechanic

  • Outer scroller .scroll-stage height set in viewport units (default 600vh = scroll 5 screens to traverse all frames).
  • Sticky hero position: sticky; top: 0; height: 100vh containing all frames absolutely positioned and hidden except the active one.
  • Scroll listener computes progress = scrolledIntoStage / scrollableHeight clamped 0–1.
  • idx = floor(progress * (N - 1)) — swap the .active class to that frame.
  • Wrap the listener in requestAnimationFrame to keep 60fps.

Design rules

  • If the project is YUV.AI-branded (Yuval personally, his portfolio, his brand): call yuv-design-system and lock it to a mode by medium —
  • Neon mode (DEFAULT for YUV.AI web): hot pink (#FF1464) + neon cyan (#00E5FF) + white or rich black canvas. Anton + Inter (English) / Rubik + Assistant (Hebrew). 0px or 999px radii.
  • Warm Editorial mode: only for Hope, Marcus, bigcats.ai, practical.yuv.ai — pink + yellow + bone, off-white never pure #fff.
  • Never use Decks mode (Fly High purple/yellow) for a landing page — that palette is reserved for slide decks only.
  • If the project is not YUV.AI-branded: do not apply yuv-design-system. Pick a palette that fits the source video's mood. Anton + Inter remain reasonable defaults but aren't mandatory.
  • Hero overlays should be subtle — a vignette and a single big headline; don't drown the frame.
  • Below-hero sections: light canvas (white in Neon mode, off-white in Warm Editorial, bone if going generic-warm), large Anton heading, body copy in Inter at 18–22px, CTA as a sharp pill (border-radius: 999px) — pink fill in Neon mode, brand-appropriate color otherwise.
  • For Hebrew: respect RTL, use Rubik 900 for headlines, Assistant 400/500 for body.

File references

FilePurpose
references/extract-frames.pyffmpeg-based frame extractor; optionally builds the full HTML in one go
references/landing-template.htmlSelf-contained landing-page skeleton with the scroll mechanic
references/design-notes.mdTypography, palette, and motion defaults

Verwandte Aufgaben

aus demselben Repository

Weitere Skills

Alle Skills
hoodini
Community

aws-account-management

Manage AWS accounts, organizations, IAM, and billing. Use when setting up AWS Organizations, managing IAM policies, controlling costs, or implementing multi-account strategies. Triggers on AWS Organizations, AWS IAM, AWS billing, Cost Explorer, SCPs, multi-account, AWS SSO, Identity Center.

Installationen
1
GitHub Stars
287
Aktualisiert
11. Juli
hoodini
Community

aws-harness

Build a new AI agent on AWS and deploy it easily, OR wrap and deploy an agent you already have, using the Amazon Bedrock AgentCore harness. Explains what an "agent harness" is (the runtime scaffolding around a model - agent loop, tool execution, memory, identity, observability), then gives two fully-working, verified paths - (A) scaffold + ship a new agent with the AgentCore CLI (create/dev/deploy/invoke), and (B) deploy a prepared agent (Strands, LangGraph, or custom) via the SDK wrapper or a F

Installationen
1
GitHub Stars
287
Aktualisiert
11. Juli
hoodini
Community

aws-strands

Build AI agents with the Strands Agents SDK - the open-source framework (the agent "brain") for writing agent logic, tools, and multi-agent systems in Python. Model-agnostic, AWS Bedrock by default. Covers Agent, the @tool decorator, model providers (BedrockModel), multi-agent patterns (agents-as-tools, Swarm, Graph), conversation management, and streaming. Every import verified against official Strands docs. To DEPLOY a Strands agent on AWS, use the aws-harness skill. Triggers on Strands, Stran

Installationen
1
GitHub Stars
287
Aktualisiert
11. Juli
hoodini
Community

bun

Build fast applications with Bun JavaScript runtime. Use when creating Bun projects, using Bun APIs, bundling, testing, or optimizing Node.js alternatives. Triggers on Bun, Bun runtime, bun.sh, bunx, Bun serve, Bun test, JavaScript runtime.

Installationen
1
GitHub Stars
287
Aktualisiert
11. Juli