// open source · claude code skill · local-first

Visual Study Atlas

Claude Code skillOffline once builtMIT licenceQuizzes · Diagrams · Flashcards

Turn a book, paper, lecture transcript, notes or a code repository into a diagram-first study atlas: one page per chapter, hand-built diagrams, definition tables, a coverage checklist, scored quizzes, flashcards for Anki, a mind map, and an ask box that runs on a model on your own computer and cites the page of your source.

▶ Watch the walkthrough Open the live demo atlas ⌥ GitHub

Full walkthrough (8.5 min)

Why it exists, what you need (including Ollama), a full install from scratch with real terminal output, the first run, and every feature of the finished atlas.

The whole video uses a text-to-speech voice. Terminal scenes are real output captured from a clean folder; app scenes are real screen captures of the demo atlas. Open on YouTube ↗

Why it exists

Three problems with studying a long, dense book, and what the atlas does about each.

Missed concepts

You finish a chapter and still miss things.

Fix: a coverage checklist closes every section, so you can see nothing was skipped.

Passive reading

Reading does not show whether you can answer in your test's style.

Fix: scored quizzes in a style you choose once for the whole project.

Getting stuck

You leave the page to search and lose the thread.

Fix: an ask box that answers from your own source, privately, with the page cited.

What you get

Everything except the pilot is optional.

Pilot first

One small chapter plus a questionnaire (max 25 questions) before the full build; your answers are frozen as the spec.

Hand-built diagrams

Pipelines, loops, architectures and charts drawn to scale; an overflow checker catches spilled labels.

Coverage checklist

Closes every section; anything outside the source is flagged "Beyond the book: verify".

Scored quizzes

Exam-style scenarios, plain recall or true/false; instant explanation, score, retry missed.

Local ask box

Click any sentence or diagram box and ask. Runs on Ollama on your machine; answers cite Textbook p.N.

Source page links

Every section links to the page of your own PDF where it starts, with sub-topic chips.

Anki flashcards

Definitions, self-checks and quiz items exported as a file Anki imports; Anki handles spaced repetition.

Mind map

Built from the sections and checklists; click a section to see its checklist.

Step-through diagrams

Flow diagrams can be stepped through one box at a time, with a caption per step.

See it

Screens from the demo atlas, built from an original three-page text about home budgeting. Open the live, read-only copy →

A diagram section with a link to the source page — One claim, one picture, a link to your source page

A flow diagram being stepped through — Step through a flow diagram, one box at a time

A scored scenario quiz with explanations — Scored scenario quizzes with explanations

The ask box answering with a cited source page — Local ask box; the answer cites the source page

The mind map page — Mind map built from the same structure

How it works: two passes

The skill stops after the pilot so you can steer before the full build.

Read. It reads your material and learns its structure.
Pilot. It picks one small chapter at random, builds it, and writes a questionnaire workbook (at most 25 questions).
You answer. Purpose, depth, quiz style, how you want it delivered. Then it stops.
Freeze. Your answers become the specification; nothing changes without asking you.
Build. The full atlas, chapter by chapter, every diagram checked.
Deliver. A local folder you open in the browser (double-click the launcher and keep that window open).

What you need

Tested on macOS only.

Need	Why
Claude Code	Runs the skill (installed as a plugin).
Python 3.9+	With `openpyxl` and `pypdf`; the skill's scripts use them.
Your source	For page links and cited answers it must be a PDF with a text layer, not a scan.
A web browser	To read the atlas.
Ollama (optional)	Runs the local models behind the ask box. Used here: `qwen3.5:9b` (6.6 GB, quick), `gpt-oss:20b` (13 GB, default and deeper), `nomic-embed-text` (274 MB, embeddings). Set the names in `config.json`; bigger models answer better but need more memory and are slower.

Install and try it

Inside Claude Code:

/plugin marketplace add sabya-chow/visual-study-atlas
/plugin install visual-study-atlas@visual-study-atlas-marketplace

Packages and the demo:

python3 -m venv .venv && source .venv/bin/activate
pip install openpyxl pypdf
git clone https://github.com/sabya-chow/visual-study-atlas
cd visual-study-atlas/examples/demo/app && python3 server.py   # then open http://127.0.0.1:8801

Then run /visual-study-atlas on your own material. Full steps with real output are in the video above and in the README.

Honest limits, copyright and licence

Local models can be wrong. Check facts against your source.
Page links land on the page where a heading starts, not on the paragraph, and need a PDF with a text layer.
Reference videos are picks by title and channel, not watched.
Step-through suits flow diagrams, not tables. A full build takes time and Claude usage, so start with the pilot.
Copyright. The repository holds tools and original demo content only. Use sources you are allowed to use and do not publish atlases built from copyrighted books. Not affiliated with any exam body, publisher or model vendor. Code is MIT licensed; the demo content is CC0.

Academic project disclaimer: This website and the projects linked here are academic and personal learning projects only. Views expressed are strictly personal and are not to be construed as the views, policy, endorsement, or advice of my organization or employer.

⌥ Repository ▷ YouTube Live demo atlas ← Back to home