raiyanyahya/Superlearn: Learn anything, deeply from inside Claude Code

Learn anything, deeply from inside Claude Code

Summary

Superlearn is a plugin and Agent Skill for Claude Code, OpenAI Codex, and Kilo Code that turns a topic prompt into a locally served interactive learning board. The agent researches the web (arXiv papers, YouTube videos), authors structured notes, and serves the result as a single-page web app. Everything is stored as plain JSON on disk, and no extra API keys are required.

What it is useful for

Engineers who want to dig into a technical subject without leaving their coding agent. Documented examples include Postgres internals, transformer neural networks, Rust ownership, React hooks, and diffusion models. Boards support six layouts (Board, Notes, Grid, Mindmap, Canvas whiteboard, Feed) and four modes: study (default), interview (likely questions and live-coding katas), research (literature map), and documentation (code-first reference). Cards include runnable code (JavaScript and HTML instantly; Python via Pyodide with NumPy, pandas, matplotlib, SymPy), Mermaid diagrams, KaTeX math, SM-2 spaced-repetition flashcards, and per-card annotations and highlights that persist in the board JSON. Themes include blueprint (engineering), terminal (systems), sepia (philosophy), and three others.

How engineers can use it

Requirements: Claude Code, OpenAI Codex, or Kilo, plus Python 3.9+. The README states scripts use only the Python standard library.

Claude Code — install from the marketplace:

/plugin marketplace add raiyanyahya/superlearn
/plugin install superlearn@superlearn-marketplace

Or from a clone: git clone https://github.com/raiyanyahya/superlearn then claude --plugin-dir /path/to/superlearn.

OpenAI Codex or Kilo:

git clone https://github.com/raiyanyahya/superlearn
python3 superlearn/scripts/install.py --codex      # and/or --kilocode

Codex users must add network_access = true under [sandbox_workspace_write] in ~/.codex/config.toml, because the scrapers need network access. Kilo needs web search enabled under Settings → Web Tools for non-default providers.

Invocation examples from the README:

/superlearn how Postgres internals work
/superlearn rust ownership — I already know C++
/superlearn react hooks for my frontend interview
/superlearn diffusion models, research mode

The local app runs at http://localhost:4321. The session stays live, so follow-up prompts like "go deeper on X" or "add the original papers" hot-reload the page with "Updated" badges.

To explore without researching, serve the shipped example:

python3 scripts/serve.py --boards-dir examples --port 4321

Export any board to a single standalone HTML file with python3 scripts/export_html.py. Pass --offline to inline Mermaid, KaTeX, highlight.js, and figures so the file works without a network connection.

Documented limitations: Runnable code is treated as untrusted and runs in a sandboxed iframe without allow-same-origin; blocks that cannot run in a browser (e.g., PyTorch examples) are marked runnable: false and do not offer a Run button. Research content belongs to its original sources, so Superlearn cites rather than copies, plays videos through YouTube's embedded player, and pulls papers from the official arXiv API. The server binds to loopback by default and validates the Host header to block DNS rebinding. The project is MIT-licensed.