write-huashu-design

Generate HTML-based prototypes, slide decks, animations, and infographics with export to MP4, GIF, PDF, and PPTX.

Updated Aug 10, 2026
One-click install
npx skills add https://github.com/Choi-Keith/skill-arsenal-ultra --skill write-huashu-design-choi-keith
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: write-huashu-design
Source: https://github.com/Choi-Keith/skill-arsenal-ultra/tree/main/skills/write-huashu-design
Command: npx skills add https://github.com/Choi-Keith/skill-arsenal-ultra --skill write-huashu-design-choi-keith

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires playwright, sharp, pptxgenjs, pdf-lib, and includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve? AI agents produce generic, low-quality visual output when asked to design prototypes, presentations, or animations. This Skill encodes a designer's workflow—brand asset protocols, anti-AI-slop rules, three-direction exploration, and staged checkpoints—so agents deliver polished HTML designs, motion graphics, and editable decks instead of purple-gradient filler. ## Core Features & Use Cases - Interactive Prototypes: Build clickable single-file HTML App/Web mockups with accurate iPhone 15 Pro bezels, real imagery, and Playwright-verified interactions. - Slide Decks & PPTX: Produce 1920×1080 HTML decks with an overview wall, then export to vector PDF or genuinely editable PPTX via an html2pptx element translator. - Motion Design & Video: Create timeline-driven animations (Stage + Sprite model) and export MP4/GIF with 60fps interpolation, BGM, and SFX; optional TTS narration pipeline for long explainer videos. - Design Direction Advisor: For vague briefs, run three parallel logic tracks (random style wheel, award-winning reference migration, top-studio philosophy) to generate three real visual drafts for user selection. - Use Case: Ask the agent to "make a 60-second product launch animation for my app"—it verifies product facts via web search, collects brand assets, presents three direction boards, then renders a scored MP4 with music and sound effects. ## Quick Start Ask your agent to make a clickable iOS prototype for a Pomodoro app with four screens, or to turn a concept into a 60-second animation exported as MP4 and GIF.

Frequently Asked Questions about write-huashu-design

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I create a clickable app prototype with an AI agent?

Describe the app and its core screens; the skill builds a single-file HTML prototype with an accurate iPhone 15 Pro frame, state-driven navigation, and real images from Wikimedia or Unsplash. Playwright runs click tests before delivery to catch interaction bugs.

How to convert HTML slides to editable PowerPoint?

The html2pptx.js script reads DOM computed styles and translates each element into native PowerPoint objects, producing real editable text frames rather than embedded images. The HTML must satisfy four structural constraints documented in references/editable-pptx.md.

Can this skill export animations as MP4 or GIF?

Yes. render-video.js records 25fps MP4 from animation HTML, convert-formats.sh derives 60fps interpolated MP4 and palette-optimized GIF, and add-music.sh mixes in background music. A seek-based renderer provides deterministic 60fps output for Stage-clock animations.

Does huashu-design work with Cursor or Codex, not just Claude Code?

Yes, it is agent-agnostic and works with any markdown-skill-capable agent including Cursor, Codex, Trae, and OpenClaw. Runtimes without subagent support run the three design directions serially instead of in parallel.

What are the limitations of HTML-based design generation?

Output is HTML, so there is no layer-editable round-trip into Figma or Keynote. Framer-Motion-tier 3D, physics, and particle animations are out of scope, and brand-from-scratch designs without any provided assets drop noticeably in quality.

Does the skill send my data to external servers?

The core pipeline runs fully locally with zero network calls. Optional cloud features (Doubao TTS, AI video review) use your own API keys, call only official vendor endpoints, and require an explicit --yes consent flag before any data leaves your machine.