← all repositories
alchaincyf/huashu-design

Turn a sentence into a launch video, prototype, or deck

huashu-design is an agent skill that converts a brief sentence into a finished HTML prototype, MP4 animation, or editable PowerPoint.

huashu-design
Velocity · 7d
+406
★ / day
Trend
accelerating
star history

What it does

huashu-design is a markdown skill that drops into Claude Code, Cursor, or any agent that reads skill files. You describe a deliverable—an iOS prototype, a launch animation, a slide deck, or an infographic—and the agent returns a polished HTML file, often exported to MP4, GIF, editable PPTX, or PDF. The built-in rubric aims for “real design team” fidelity rather than the usual purple-gradient AI slop.

The interesting bit

Instead of heroic one-shot generation, the skill enforces a “Junior Designer Workflow”: the agent must show early gray-block drafts, list assumptions, and iterate before finalizing. For any branded task, a mandatory five-step “Core Asset Protocol” makes it hunt down actual logos, product shots, and UI screenshots rather than guessing hex codes from memory. It even runs a self-critique across five dimensions—philosophical coherence, hierarchy, craft, functionality, innovation—before signing off.

Key highlights

  • Agent-agnostic: works with Claude Code, Cursor, Codex, Hermes, OpenClaw, and other markdown-skill-capable agents
  • Deliverables include clickable HTML prototypes with Playwright-verified interactions, MP4/GIF motion graphics, and PPTX slides with real text frames
  • 20 built-in design philosophies and explicit “anti AI-slop” rules (e.g., no CSS silhouettes standing in for product shots, no Inter-as-display default)
  • Rigid brand-asset pipeline: searches official press pages, downloads and verifies images, then freezes everything into a brand-spec.md before designing
  • 5-dimension expert critique mode scores work and outputs a Keep/Fix/Quick Wins punch list with radar-chart visualization

Caveats

  • The skill’s internal prompts and reference files are written in Chinese; the agent handles English tasks, but the underlying rubric is natively bilingual by necessity
  • Interaction is purely conversational—type a sentence and wait—so there is no GUI canvas or click-and-drag editing
  • Brand work depends on the agent successfully executing web searches and downloads; if the search fails, the “never guess from memory” rule can block progress

Verdict

Great for developers who want presentable mocks, promo clips, or pitch decks without leaving their agent chat. Less useful if you live in Figma and prefer manual pixel control.

Frequently asked

What is alchaincyf/huashu-design?
huashu-design is an agent skill that converts a brief sentence into a finished HTML prototype, MP4 animation, or editable PowerPoint.
Is huashu-design open source?
Yes — alchaincyf/huashu-design is open source, released under the MIT license.
What language is huashu-design written in?
alchaincyf/huashu-design is primarily written in HTML.
How popular is huashu-design?
alchaincyf/huashu-design has 21.8k stars on GitHub and is currently accelerating.
Where can I find huashu-design?
alchaincyf/huashu-design is on GitHub at https://github.com/alchaincyf/huashu-design.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.