Domain learn

Skill shangjunyang1986/ai-learning-skills/skills/domain-learn

Turn a GitHub project / paper / technical book / field into an offline, shareable, interactive learning page — a family of standard skills (Claude Code, Codex, Gemini, OpenCode…). Personal use; not token-optimized.

Install

npx -y skills add shangjunyang1986/ai-learning-skills --skill domain-learn

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Research an open topic / field across the web and turn it into an offline, visual, beginner-to-hands-on learning page — a single self-contained HTML file with a left table of contents, a learning roadmap, sourced explanations, the field's real images and videos downloaded locally, and (when the topic admits it) an interactive in-browser demo that teaches the core intuition. Use this whenever the user wants to LEARN, STUDY, RESEARCH, "get into", or "做个从入门到实践的教程" for a TOPIC or FIELD that is NOT a single GitHub repo — e.g. "我想学习 3DGS / 帮我研究下扩散模型 / 给我做个 RAG 从入门到实践的学习教程 / I want to get up to speed on reinforcement learning / explain CRDTs end to end". Triggers on a topic name + intent to learn/understand it, even without the words "website" or "HTML". Distinguish from siblings: a GitHub URL → use github-project-learn; a specific book/PDF → use textbook-learn; a bare topic/field → THIS skill. Not for: answering a quick factual question (just answer it), or building generic web apps.

SKILL.md

5.4 KB, as published. Nobody here has run it

domain-learn

Turn an open topic (a research field, a technique, a concept) into a single, offline-openable learning page that takes someone from intuition → theory → hands-on → frontier. Left = table of contents, right = a learning roadmap and sourced content, with the field's own diagrams and videos downloaded locally and — when the topic is visual or algorithmic — one interactive in-browser demo that makes the core idea click.

This is the family's sibling to github-project-learn. The output shell is the same; what differs is the source (the open web, not one repo), the research rigor (everything sourced — the web is fuzzy and we must not fabricate), the pedagogy (a beginner→frontier roadmap), and the interactivity (a teaching demo, not just a code walkthrough).

When to use

The description covers triggering. In short: a topic/field name + any "I want to learn / research / understand this" intent, where the subject is not a single GitHub repo or a specific book. Examples: "学习 3DGS", "研究下扩散模型", "RAG 从入门到实践", "explain CRDTs".

Output shape

A folder (default <topic>-learn/) containing:

index.html — self-contained (inline CSS+JS), double-click to open, fully offline
assets/ — the field's real images/diagrams/video covers, downloaded

Workflow (high level)

Scope the topic with the user. Confirm the exact subject and the depth (a quick orientation vs a thorough入门→前沿 treatment). Tell them you'll build a single self-contained offline page.
Research rigorously, in parallel. Fan out (subagents when available) over three angles: (a) fundamentals + the seminal work, (b) the ecosystem / tools / key implementations, (c) learning resources + videos + a suggested path. Every non-obvious claim, number, and link must come from a real fetch and carry a source. Flag anything you can't verify rather than guessing. See references/research-sourcing.md — this is the make-or-break difference for an open topic.
Design the learning roadmap (入门 → 进阶 → 实践 → 前沿) and map resources to each stage.
Build one interactive demo if the topic admits a clean visual/algorithmic intuition — a self-contained, offline canvas/SVG/WebGL playground with controls that teaches ONE core concept. If the topic has no clean visual, skip it rather than forcing one. See references/interactive-demos.md.
Download media with scripts/fetch-media.sh into assets/ (handles Git LFS and blocked hosts; drops broken/placeholder covers automatically).
Generate the page by copying assets/template.html (a complete worked example — the 3D Gaussian Splatting page) and replacing its content, keeping the CSS, the JavaScript (lightbox, scroll-spy, glossary search, copy buttons, video cards), and the section scaffolding.

Read references/workflow.md for step-by-step detail, references/page-design.md for the section list + components, references/research-sourcing.md before researching, and references/interactive-demos.md before building the demo.

Bundled resources

assets/template.html — the proven page (a worked 3D Gaussian Splatting example). Copy it, keep its CSS + JS (lightbox, scroll-spy, searchable glossary, copy buttons, video cards, and the interactive-demo pattern), replace all content.
scripts/fetch-media.sh — ./fetch-media.sh <out>/assets name1=url1 …. Auto-fixes Git LFS pointers, routes blocked hosts through a proxy, and drops invalid/placeholder images. Run via the Bash tool (git-bash curl), not Windows cmd curl.

Principles that make the page good

Faithful, never fabricated — and sourced. Open-web topics are full of half-truths. Cite the paper / official page / API for every claim; flag uncertainty; say "no good resource exists" when true. This is the skill's single most important rule.
Roadmap, not a dump. Organize for a learner's journey (intuition first, math second, hands-on third, frontier last), not as an encyclopedia.
One great demo beats ten paragraphs. When the topic is visual/algorithmic, an interactive playground is the highest-value thing on the page.
Offline-first. Relative assets/ paths; verify every download is a real image.
Confirm scope before a long research run.