Duolingo Learning OS

Setup

On first use, read setup.md for routing, filesystem bootstrap, and topic activation.

When to Use

User wants to learn one or more subjects with short daily lessons, instant feedback, streak pressure, and ongoing review. Agent runs a Duolingo-style operating system (not just advice) with topic folders, per-topic curriculum, cross-topic planning, and optional AGENTS router integration.

Architecture

State lives in ~/duolingo/. See memory-template.md for global state and per-topic file templates.

~/duolingo/
|-- memory.md                    # Global status and active topic map
|-- router/
|   |-- topics.md                # Canonical list of active topics and trigger phrases
|   `-- agentsmd-snippet.md      # Snippet to keep AGENTS routing in sync
|-- topics/
|   |-- english/
|   |   |-- profile.md           # Goal, level, pace, constraints
|   |   |-- curriculum.md        # Skill tree and lesson units
|   |   |-- queue.md             # Next lessons and review backlog
|   |   |-- sessions.md          # Session history and outcomes
|   |   `-- checkpoints.md       # Weekly and milestone checks
|   `-- cooking/                 # Same structure for any other topic
`-- archive/                     # Retired topics and old curriculum versions

Quick Reference

Core Rules

1. Bootstrap the Learning OS Before First Lesson

If ~/duolingo/ is missing or empty, create the full scaffold from blueprint.md. Do not start teaching without:

• global memory file
• router files
• at least one topic namespace

2. Offer Optional AGENTS Router Integration

If the user wants automatic routing, provide the AGENTS router snippet from activation-routing.md. Never auto-edit AGENTS from this skill. Generate snippet text only and let the user apply it. If the user skips router integration, keep the skill fully manual and user-invocable.

3. Keep Each Topic Isolated, Then Coordinate Globally

Every topic gets its own namespace under ~/duolingo/topics/<topic-slug>/. Never mix files across topics. Global planning may schedule multiple topics in one day, but lesson state stays per-topic.

4. Run Lessons as Tight Interactive Loops

Use lesson-loop.md for every session:

1. micro-challenge
2. learner attempt
3. immediate correction
4. reinforcement challenge
5. queue update

One loop should finish in about 60-90 seconds.