这是一项Claude技能,构建精致的互动HTML解释器,配备自动QA循环和防垃圾视觉保护。
概览
一个Claude,用来构建精致、互动的HTML解释,既可以是单页,也可以是幻灯片。给它一个主题,回答几个问题,它就能生成一个真正让人摸索、理解基础的页面。 它是一个:质量来自一个迭代循环(学习时研究架构师→×视觉设计→→渲染和实际呈现→修改),而不是填补空白。 它既更快,质量也更高,因为更强的初稿意味着更少的质量循环修复周期。Claude Code 在真正的无头浏览器中渲染,因此能捕捉 Claude App 或 Cowork 无法捕捉的视觉错误。 给定一个主题(或你自己的素材),技能是: - — 一个简短、必问的访谈:基础、受众深度、格式(滚动页面与幻灯片)、视觉风格、测验、AI聊天。- — 基于你的材料和/或原始资料;引用它们。- — 以一个心理模型开头,层叠深度,围绕一个清晰的核心构建。
概述(AI 翻译)
一个Claude 技能,用于构建精致、互动的HTML解释,既能做成单页,也可以做幻灯片。给它一个主题,回答几个问题,它就生成了一个真实的页面,用户可以冷静地看到它,并且理解了基础知识。
这是一个过程,不是模板:质量来自一个迭代循环(研究→ 学习建筑师 → 10×视觉设计→渲染和实际呈现→修改),不是 从填补空白开始。
观看实际操作:现场示例图库
最佳效果:用Claude Code搭配Opus 4.8运行,且工作量更高(或更好)。 它既更快又质量更高,因为更强的初稿意味着更少的质量循环修复周期。Claude Code在真正的无头浏览器中渲染,因此能捕捉Claude App或Cowork无法捕捉的视觉bug。
它的作用
给定一个主题(或你自己的素材),技能是:
- 先看范围 — 简短且必问的访谈:基础、受众深度、形式 (滚动页面与幻灯片)、视觉风格、测验、AI聊天。
- 准确事实——基于你的材料和/或原始资料;引用它们。
- 学习架构师 — 以一个心智模型为导向,层叠深度,围绕
本页由 AI 翻译,技术细节请以原文为准。
README
以下为原文 README 节选,完整内容请查看原文。
HTML Explainer
A Claude Skill for building polished, interactive HTML explainers as a single page OR a slide deck. Give it a topic, answer a few questions, and it produces a real page that someone can land on cold and come away understanding the basics.
It’s a process, not a template: the quality comes from an iterative loop (research → architect for learning → 10× the visual design → render-and-actually-look → revise), not from filling in blanks.
See it in action: live examples gallery
Best results: run it in Claude Code with Opus 4.8 at high effort (or better). It’s both faster and higher quality, since stronger first drafts mean fewer quality-loop fix cycles. Claude Code renders in a real headless browser, so it catches visual bugs the Claude App or Cowork cannot.
What it does
Given a topic (or your own source material), the skill:
- Scopes first — a short, must-ask interview: grounding, audience/depth, format (scrolling page vs slide deck), visual style, quiz, AI chat.
- Gets the facts right — grounds in your material and/or primary sources; cites them.
- Architects for learning — leads with one mental model, layers depth, builds around a single clear centerpiece.
- Hits a 10× visual bar — editorial typography, a semantic palette, real use of the horizontal space, and an explicit anti-“AI-slop” checklist (no emoji-as-icons, no decorative gradients, no generic bento grids).
- Shows, doesn’t tell — favors small playable micro-demos over paragraphs.
- Runs a real quality loop — renders the page in a headless browser, screenshots it (desktop + mobile + interactive states), critiques it like a human, and fixes what’s weak.
- Ships a real
.htmlfile with an optional in-page Review & edit overlay so non-technical reviewers can annotate or edit in place (and download a copy with both the edits and the notes embedded).
Examples
Built with this skill, hosted on GitHub Pages — each is a self-contained .html file:
| Explainer | Format | Live |
|---|---|---|
| Auroras: The Northern Lights | scrolling page | open |
| Gutenberg’s printing press | slide deck | open |
| The Doppler effect (why a siren changes pitch) | scrolling page | open |
| The grammar of shadow (reading film noir) | slide deck | open |
| The EU AI Act | scrolling page | open |
| What “fair use” actually means | slide deck | open |
推荐工具
换一个关键词,或者移除筛选条件。
安装
npx skillfish add ds-vibe/html-explainer