teach

/teach — Framework Knowledge Transfer

When to use

When user says "teach", "explain framework", "onboard", "how does this work", "tutorial", "walk me through", or wants a structured overview of the Torus framework.

Commands

• /teach — Full framework tutorial
• /teach architecture — Deep-dive on architecture only
• /teach patterns — Key patterns and conventions
• /teach pitfalls — Common mistakes and how to avoid them
• /teach $ARGUMENTS — Focus on a specific area (e.g., "gates", "memory", "agents")

Flow

Phase 1: INVENTORY

Query memory for high-value patterns:

search_knowledge("type:learning", top_k=30)
search_knowledge("type:fix", top_k=20)
search_knowledge("type:decision", top_k=20)

Count entries by category:

• Architectural decisions
• Key patterns / conventions
• Common pitfalls / fixes
• Best practices

Phase 2: CURATE

Select the 10 most important entries by:

1. Relevance to requested topic (or all topics if no argument)
2. Recency (prefer entries from last 10 sessions)
3. Tag priority: priority:critical > priority:high > priority:medium
4. Outcome: prefer outcome:success entries

For each selected entry, call get_memory(id) to retrieve full content. Note the memory ID for citation.

Phase 3: FORMAT

Structure the tutorial as markdown with these sections:

## Architecture
Overview of how the framework is organized: gates, hooks, memory, agents, skills.
Reference key files: enforcer.py, memory_server.py, CLAUDE.md.

## Key Patterns
The recurring conventions every contributor must know:
- Memory-first workflow
- The Loop (memory → plan → test → build → prove → ship)
- Gate contracts and exit codes
- Agent delegation rules

## Common Pitfalls
Mistakes that have burned us before (cited from fix memory):
- [pitfall] — [memory_id] — [how to avoid]

## Best Practices
Proven approaches (cited from decision/learning memory):
- [practice] — [memory_id] — [why it matters]

Phase 4: DELIVER

Present the tutorial in clear markdown. After each section, offer: "Want me to deep-dive on [section]?"