- Research Context — Why this project exists, landscape analysis, platform comparison
- Feature Spec — Complete feature list with rationale for each
- Architecture — System diagram, file structures, design decisions, command routing
- Curriculum — 8 modules, 40 lessons, 8 projects, concept dependency graph
- Pedagogy — Learning science, Socratic method, spaced repetition, anti-patterns
- Data Model — JSON schemas for all state files, update rules, ownership
- Option A workflow: User writes real .py files in VS Code, Claude reads + tests + gives Socratic feedback
- Never give the answer: Graduated hints, 4 levels, full explanation only as last resort
- Never ask twice:
environment.jsonstores all setup state permanently - Session bootstrap: Every new Claude conversation reads all state files before responding
- Adaptive: Comfort scores (0-100) per concept drive difficulty and review scheduling
- Memory across sessions: All progress persists in
~/learning/.claude/as JSON files