Build course
Build a complete interactive HTML course on any topic — curriculum, lessons, interactive drawings per module, mid-module quick checks, and an auto-scored final quiz with evaluation. Use whenever the user wants to learn or be taught something substantial, even if they never say "course" — "teach me X", "help me learn X", "explain X properly", "I want to understand X", "build a course/curriculum/lesson on X", "make study material for X", or any request for educational content as HTML. Do not use for a quick factual answer with no course requested, for reviewing or editing an existing course file, or for non-HTML educational content like a blog post or slide deck — use a content or review skill for those.From its SKILL.md
npx -y skills add Arigatouz/teach-me --skill build-courseAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 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.
SKILL.md
7.8 KB, as published. Nobody here has run it
Build Course
Produce a single self-contained HTML file that teaches the requested topic end-to-end: curriculum → lessons → visualizations → quick checks → final quiz → evaluation. The file skeleton and engine are shipped as tested assets — fill in content, never rebuild the machinery.
Workflow
Step 1 — Clarify (only what's missing)
Determine from the request; ask via AskUserQuestion only for what's genuinely unknown:
- Topic and scope — what exactly to teach
- Learner level — beginner / intermediate / advanced
- Course size — quick lesson (1 module, ~30 min) / standard (4–8 modules) / deep curriculum (10+ modules). Default: standard.
- Language — match the user's language unless told otherwise
- Research approach — whenever the course needs the full pipeline (standard or deep course, see
references/research-pipeline.md): always ask, in this same round, how to research it — the plugin's research swarm (default) or the user's deep-research skill if one is installed. Never skip this question and never decide it yourself.
One round of questions maximum, and AskUserQuestion caps at 4 questions per round. If more than 4 of the above are genuinely unknown, ask topic/scope and research approach first — they change what gets built — and default the rest (learner level → intermediate, course size → standard, language → the user's message language) instead of exceeding the cap. Sensible defaults beat interrogation.
Step 2 — Research (verified pipeline)
Read references/research-pipeline.md and follow it. In short: if the topic is fast-changing, has official documentation, or the user wants a deep curriculum, research is mandatory BEFORE writing content. Scale it to course size — quick lessons get lightweight inline research; standard and deep courses get the full pipeline, using whichever engine was chosen in Step 1 (always asked, never decided by you). If the swarm was chosen or defaulted to, spawn the plugin's three researcher agents in parallel (course-researcher-docs, course-researcher-pedagogy, course-researcher-ecosystem — each has a deliberately different search strategy), then the course-verifier agent, which checks technical claims against the official docs (docs MCP first, official-site fetch otherwise). Doing this research or verification yourself instead of spawning the agents is not a substitute, no matter how confident you are in a direct tool call — run the agents. Write the course only from the verifier's output, and embed the sources list in the course home section. Skip research only for stable, timeless topics at quick-lesson depth. Never invent statistics, version numbers, or dates — verify or omit.
Step 3 — Design the curriculum
Before writing HTML, draft the curriculum and show it to the user in chat. When the research pipeline ran, build the plan from research/verified.md — its teaching guidance dictates module order, and its course facts bound what can be taught:
- 3–7 learning objectives ("After this course you can…")
- Module list: title, one-line summary, estimated time
- For EACH module: which concept gets its interactive drawing, and which pattern (slider or step-through)
- Quiz plan: 2 final-quiz questions per module minimum, mixed types, plus 1–2 quick checks inside each module
Proceed immediately after showing the plan unless the user asked to review first.
Step 4 — Build from the template
Copy assets/template.html (in this skill's directory) to the outputs folder as <topic-slug>-course.html. Fill in every {{PLACEHOLDER}}:
| Placeholder | Content |
|---|---|
{{LANG}} | Page language code (en, ar, …) |
{{COURSE_TITLE}} / {{SHORT_TITLE}} | Full and sidebar titles |
{{NAV_LINKS}} | One <a href="#mN" class="nav-link" data-nav="mN"><span class="nav-num">N</span> Title</a> per module |
{{HOME_SECTION}} | h1, .lede, objectives card, module overview card |
{{MODULE_SECTIONS}} | One <section id="mN" class="module"> per module (structure below) |
{{QUIZ_COUNT_LINE}} | e.g. "10 questions · covers all modules · you can retake it" |
{{QUIZ_QUESTIONS}} | Question objects — format in references/quiz-engine.md |
{{QUICK_CHECKS}} | Quick-check objects keyed by placeholder div id — format in references/quiz-engine.md |
{{WIDGET_SCRIPTS}} | <script> blocks for interactive widgets (or empty) |
Do not modify the shipped CSS or engine scripts — only replace placeholders and add widget scripts. references/html-template.md documents the design system and available component classes.
Each module section contains, in order:
.pillwith module number + estimated time, then h2 title.callout— "Why this matters" hook (2–3 sentences)- Lesson content: short paragraphs, concrete examples, one analogy per abstract concept; code in
<pre class="code"> - EXACTLY ONE OR MORE interactive drawings — this is mandatory per module, not optional (
references/visualizations.md): a slider widget when the concept has a variable parameter, a step-through widget for sequences/algorithms, or a toggle/click-to-explore widget otherwise. Static SVG diagrams and Chart.js charts are welcome additions, but they supplement the interactive drawing, never replace it. The verifier fails any module without one. - 1–2
<div class="quickcheck" id="qcN"></div>placed right after the concept they test .callout.goodwith 3 key-takeaway bullets<button class="btn ghost mark-done" data-module="mN">Mark module complete</button>
The engine already provides: sidebar progress bar, progress + best-score persistence via localStorage, quiz shuffling on every attempt, per-topic evaluation with study recommendations.
Step 5 — Verify (mandatory, one command)
node <skill-dir>/scripts/verify.js <path-to-course.html>
It checks: no leftover placeholders, all scripts parse, nav links point at real sections, balanced/unique structure, offline safety, quiz schema, scoring math, shuffle correctness, quick-check wiring. Must print ALL CHECKS PASSED. If it fails, fix the reported issues and re-run. Never deliver a file that hasn't passed.
Step 6 — Deliver
Present the file with a one-paragraph summary: module count, visual count, quiz size. Offer follow-ups: more depth on a module, a harder quiz, or a printable PDF version.
Hard rules
- Single file. All CSS/JS inline, diagrams as inline SVG. No frameworks, no build step, no external images.
- Only permitted external resource: Chart.js from cdnjs.cloudflare.com, wrapped in an offline fallback.
- Works from
file://— no fetch(), no ES modules. - Every widget in its own IIFE with unique id prefix (
w1,w2, …). - Accessible: semantic headings, radio groups inside
<fieldset>/<legend>(the engine does this),aria-labelon controls, contrast ≥ 4.5:1. - Quiz answers: 2+ per module, all three types represented, every question has
explain. Fill-in answers list all reasonable phrasings. - The research engine (Step 1) is always asked, never silently decided — and whatever gets chosen is executed exactly as documented, not swapped for your own direct research.