Course: hardening-proof-1781911772297

---
name: visual-teach
description: Stateful, multi-session teaching workspace that grounds every lesson in the user's mission and delivers each one as a beautiful, self-contained interactive HTML lesson (warm-neutral design system with built-in dark mode, one byte-identical shared shell across all pages, inline SVG diagrams, 20 ready-made interactive demo types, a Simple↔Technical toggle, an index hub, and EN + PT-BR shipped by default). Self-grills its own teaching plan (grill-me, self-answered) before building, and **self-manages**: in autonomous mode it runs a maintenance loop — scan the workspace, do the next lesson, refresh stale ones, keep the glossary/resources/learning-records healthy, log progress, and surface only decision-ready choices. Use when the user wants to learn a topic or a codebase deeply over time, asks you to "teach me X", "build a course / visual course / interactive onboarding / didactic walkthrough", "keep teaching / keep going / run the course / maintain this", or to run an ongoing tutor that tracks what they've learned. The state (mission, resources, glossary, learning records, lessons) lives as files in the current directory. The skill is fully self-contained — design system, all 20 demos, diagram recipes, teaching language, and pedagogy ship inside it; no other skill is needed. NOT for a single one-off report/spec/status doc (use visual-docs), and NOT for a quick one-line explanation in chat.
---

# Visual Teach

## Overview
A stateful tutor: treat the current directory as a **teaching workspace**. Ground all teaching in the user's **mission**, teach at the edge of their **zone of proximal development**, and deliver each lesson as a **self-contained interactive HTML file** built from the shared design system (`assets/lesson-template.html`) — inline SVG, one of 20 ready-made interactive demo types (`assets/demos/`), a Simple↔Technical toggle, and citations. **Self-grill** the teaching plan before building (`grill.md` — question and self-answer it; only genuinely user-only questions reach the user, already answered with a recommendation). Track what the user has actually learned so the next session builds on it. Everything the skill needs ships inside it — never reach for another skill.

## When to Use
- The user wants to learn a topic, skill, or codebase **deeply over multiple sessions**, or asks you to be their ongoing tutor.
- The user asks to "teach me X", "build a course / visual course / interactive onboarding / didactic walkthrough" for a system, with diagrams + interactive demos for a mixed (technical + non-technical) audience.
- When NOT to use: a single report/spec/status/post-mortem → `visual-docs`; a one-line answer the user wants right now in chat → just answer.

## The Workspace (state lives in files)
- `MISSION.md` — the *reason* the user is learning this; grounds every teaching decision. Format: `MISSION-FORMAT.md`.
- `RESOURCES.md` — curated high-trust sources (Knowledge) and communities (Wisdom). Format: `RESOURCES-FORMAT.md`.
- `GLOSSARY.md` — the canonical, opinionated language of the workspace; adhere to it in every lesson. Format: `GLOSSARY-FORMAT.md`.
- `learning-records/NNNN-slug.md` — ADR-style records of what the user has genuinely learned; drive the zone of proximal development. Format: `LEARNING-RECORD-FORMAT.md`.
- `index.html` — the course hub (from `assets/index-template.html`); the visual map of the curriculum, organized around the mission.
- `lessons/NNNN-slug.html` — the lessons, one self-contained HTML file each, built from `assets/lesson-template.html`. The primary unit of teaching.
- `reference/*.html` — compressed reusable reference docs (cheat sheets, algorithms, syntax, poses, glossary-as-page) built with the same design system; lessons link to these and they get revisited. Make them the **compressed essence** — Tufte-clean, high-density, built to last and print/export well (lessons are rarely revisited; reference docs are). The full rationale is in `references/pedagogy.md`.
- `NOTES.md` — scratchpad for the user's stated teaching preferences and your working notes.
- `MAINTENANCE.md` — the dated, high-level log of the self-maintenance loop (lessons built/refreshed, learning-records written, glossary/resource/mission changes, learner decisions, what's queued next). Lets a later session resume the loop. See `references/orchestration.md`.

## Philosophy
Deep learning needs three things: **Knowledge** (from high-trust resources — never your parametric memory), **Skills** (built by interactive, effortful practice), and **Wisdom** (from real-world practice in a community). Some topics lean knowledge (theory), others skills (yoga, coding).
Split **fluency strength** (in-the-moment recall, which feels like mastery) from **storage strength** (durable retention, the real goal). Build storage strength with desirable difficulty: **retrieval practice** (recall from memory), **spacing** (distribute over time), and **interleaving** (mix related topics —