learn-skill

A Claude Code skill that turns any subject into a science-backed learning system — Anki cards, Obsidian notes, and a spaced repetition schedule, in one command.

---

What it does

You give Claude any learning material — a PDF, a concept, a vocabulary list — and learn-skill builds a complete learning system around it:

Output	Description
Adaptive learning plan	Detects content type, asks 2-3 targeted questions, generates a session plan calibrated to your available time
Anki deck (.apkg)	Exports a ready-to-import Anki file with Basic, Cloze, and reverse cards — no manual work
Obsidian note	Structured note with concept map, active recall checklist, Feynman section, and review schedule
Spaced repetition schedule	SM-2 review dates (J+1, J+6, J+14, J+30, J+60) baked into the Obsidian note

Works for: courses & text, abstract concepts, foreign languages — in any language.

---

Demo

/learn-anything

> I pasted my thermodynamics chapter (12 pages)

Claude detects: Type A — Text/Course
Questions:
  1. Objectif : comprendre en profondeur ou mémoriser ?  →  Comprendre
  2. Temps disponible aujourd'hui ?  →  1h30
  3. Flashcards Anki ?  →  Oui

Generating...
✓ Concept map (8 core concepts)
✓ Full session plan (3 × 50 min Pomodoros)
✓ Feynman checkpoint for entropy and enthalpy
✓ 34 Anki cards → thermodynamics.apkg
✓ Obsidian note → thermodynamics.md (with review schedule)

Next review: Tomorrow, 15 min active recall only.

---

Science

This skill implements five evidence-based learning techniques:

1. Spaced Repetition (SM-2 algorithm)

Each card gets review intervals that expand as recall improves: 1d → 6d → 14d → 30d → 60d. The ease factor (default 2.5) adjusts per card based on recall quality.

2. Active Recall (Testing Effect)

Every session ends with retrieval practice — no passive re-reading. Retrieving information strengthens the memory trace 2–3× more than reviewing the source material.

3. Feynman Technique

After encoding, Claude prompts you to explain the concept in plain language. The points where you struggle are your exact knowledge gaps — those gaps generate targeted Anki cards.

4. Chunking

Content is broken into groups of 4±1 concepts (Cowan's updated model of working memory), reducing cognitive load during encoding.

5. Desirable Difficulties

The skill forces the generation effect: you attempt to recall before seeing the answer. This harder retrieval creates stronger, more durable encoding.

---

Installation

curl -fsSL https://raw.githubusercontent.com/guepardlover77/learn-skill/main/install.sh | bash

Or manually:

git clone https://github.com/guepardlover77/learn-skill
cd learn-skill
./install.sh

Then restart Claude Code. The skill activates automatically when you say "I want to learn X" or via /learn-anything.

---

Usage

Trigger phrases (any language)

• /learn-anything
• "Je veux apprendre [X]"
• "Help me memorize / understand / study [X]"
• "Generate Anki cards for [X]"

Workflow

1. Paste your material (text, notes, vocabulary list) or describe a concept
2. Answer 2-3 questions (goal, time available, Anki yes/no)
3. Get your learning system: plan + Anki script + Obsidian note
4. Run the Python script to export your .apkg Anki deck
5. Follow the review schedule — Claude tracks it in Obsidian

---

Examples

See examples/ for real session outputs:

• physics-thermodynamics.md — 12-page course → 34 Anki cards
• japanese-n5-vocab.md — 50 words → bidirectional deck
• quantum-computing.md — abstract concept → concept map + Feynman

---

Requirements

• Claude Code (any plan)
• Python 3.8+ with genanki for Anki export: pip install genanki
• Obsidian (optional, for progress tracking)
• Anki (optional, for spaced repetition)

---

How it works

The skill is a single SKILL.md file that Claude Code reads as instructions. It contains:

1. A content-type detector (text/course vs abstract concept vs language)
2. An adaptive interview (2-3 questions max, tuned per content type)
3. A session planner (sprint / full / deep-dive based on available time)
4. A card generator (Basic, Cloze, Reverse — genanki-compatible Python output)
5. An Obsidian template (concept map, Feynman section, review log)
6. A post-session protocol (what to do in the next 24h)

---

Contributing

PRs welcome. Especially interested in:

• New content-type detectors (e.g. code/programming, music theory)
• Better card templates for specific domains
• Anki-Connect integration (auto-import without the Python script)
• MCP server version

---

License

MIT — use it, fork it, improve it.