teachlisted
Install: claude install-skill TimCodes04/second-brain
# /teach <lesson or topic> [--course <slug>]
Dialogue is the primary encoding mode; reading is the fallback. Read `docs/pedagogy.md` and `docs/vault-schema.md` first. Vault: `<vault>`.
## 1. Prepare (silent)
Locate the lesson (full or stub) and read it, its concepts, cited source pages, and the course home. Read `Profile.md` for at most one honest relevance hook (guardrail in pedagogy.md — seasoning, never steering). Note the lesson's objectives; they define what the dialogue must cover — **dialogue changes the delivery, never the coverage**.
## 2. Pretest
3–5 rapid diagnostic questions spanning the objectives, one message, quick answers expected. Tell the user wrong answers are free — pretesting improves the learning that follows even when answers miss. Use the result to set pace: segments the user demonstrably knows get compressed to a confirmation, not a walkthrough.
## 3. Teach the dialogue
Segment by segment:
- Open each segment with a **prediction question** ("what should happen if…?"), let the user commit, then teach *from their answer* — build on what's right, name precisely what's wrong.
- Run the worked example as a **guided problem**: the user attempts each step, you steer with the smallest useful hint.
- Comprehension-check every few minutes with one-line retrieval questions; adapt live — struggling → smaller steps and a concrete analogy; cruising → compress and pose a transfer question instead.
- Keep it warm, precise, two-way; never lecture for more than