Anything

Learn Anything — study-app generator

Turn any subject into a polished, offline study+quiz web app. The look, code, and interaction are a fixed template; you generate only the content (the curriculum, study guides, and questions) and seed it into SQLite via the build script. Never rewrite the template's HTML/CSS/JS — drive it entirely through study.config.json + content/.

The subject to build for is: $ARGUMENTS (if empty, ask the user what they want to study).

Prerequisites

• Node ≥ 22 (the build uses the built-in node:sqlite). Check with

node --version. No npm install is ever needed — everything is vendored.

Workflow

1. Scope it

Confirm the subject and pick a size. Don't over-ask — infer sensibly, state your plan in one line, and proceed. Sizing (see references/config-schema.md):

• A real certification/exam → 12–20 topics, 30–50 questions each, exam size

matching the real test.

• A general subject → 6–12 topics, 15–30 questions each.
• A focused topic → 3–6 topics, 8–15 questions each.

2. Scaffold the project

Choose a target directory (default ./ in the current working directory, where `` is a kebab-case of the subject). Then copy the template:

mkdir -p ""
cp -R "${CLAUDE_SKILL_DIR}/template/." "/"

3. Write study.config.json

Read ${CLAUDE_SKILL_DIR}/references/config-schema.md and write /study.config.json. Design:

• domains that partition the subject, with exam weights.
• topics (id, slug, name, domain) covering the subject with no big gaps.
• levels (ideally 10) and badges with names/icons that fit the subject's

world — not generic, not AWS-flavored. Reuse the badge types from the schema.

• realistic passMark, exam.questions, exam.minutes.

4. Generate the content (parallelize)

Read ${CLAUDE_SKILL_DIR}/references/content-format.md. For every topic create, under /content//:

• questions.json — the multiple-choice questions
• guide.md — a focused Markdown study guide
• 3 .svg diagram files referenced from guide.md — each guide has exactly

3 inlined diagrams (dark-theme SVGs per the content-format spec)

Generate topics in parallel: launch subagents (Task tool) — one per topic — each writing that topic's files directly, following the content format exactly. Instruct each subagent to make questions accurate, unambiguous, position-independent (options get shuffled) with plausible distractors and 1–3 sentence explanations; a focused study guide; and 3 explanatory, self-contained dark-theme SVG diagrams (valid viewBox, light text + accent palette) that each illustrate a different concept and are referenced at the right spots in the guide. Prefer the strongest model available for content quality.

5. Build the database

cd "" && node build/build.mjs

This seeds SQLite from the config + content and emits the embedded bundles. Fix any warnings it prints (bad/missing questions, unknown domains).

6. Verify and hand off

• Open /index.html directly (works over file://), or serve it:

cd "" && python3 -m http.server 8000.

• Optionally git init the project so the user has a repo per subject.
• Tell the user: what was built (N topics, M questions), how to open it, that the

⚙ gear in the top bar switches among 14 themes (Monokai, Nord, Dracula, Tokyo Night, Catppuccin, Gruvbox, Everforest, Rosé Pine, One Dark, Kanagawa, Light, Solarized Light, …), that node build/build.mjs rebuilds after editing content, and that node update.mjs pulls the latest app template.

Rules

• Do not touch index.html, css/, js/, vendor/, or build/ in the

template — they are subject-agnostic. All customization is config + content.

• Options are shuffled at runtime; never rely on answer position.
• Guides are self-contained Markdown with 3 inline SVG diagrams each (no

raster/external images).

• Keep one emoji for icon; give levels/badges subject-appropriate names.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: makeit-run
• Source: makeit-run/learn-anything
• License: MIT

Install and usage instructions live in the source repository linked above.