Teach Skill FAQ: Personal AI Teaching With Claude Code
Frequently Asked Questions
What exactly does the Teach skill do?
How is it different from just asking Claude Code to explain something?
Where do the lessons and notes actually live?
Does it work for non-coding topics?
How does it know what to teach me next?
Why does the skill tell the AI to never trust its parametric knowledge?
Do I need to join a community to use it?
Can I adapt or reuse the generated lessons?
Is Teach better than Anki or flashcards for retention?
How do I install it, and is it free?
Why a Teach Skill FAQ Exists#
Matt Pocock’s Teach skill sits at #58 on the skills.sh all-time leaderboard with 459,800 installs, yet most coverage of the mattpocock/skills family fixes on the engineering skills — tickets, triage, ADRs — and skips this one entirely. That’s a miss, because Teach is arguably the least engineering-shaped skill in the repo, and its actual behavior is easy to guess wrong.
I wrote these answers after pulling the skill folder out of the mattpocock/skills repository and reading SKILL.md plus its five format files line by line. Everything below traces to something written in the spec — no hearsay, no marketing copy.
The Short Answers#
- It’s a teaching workspace, not a chatbot. Files, not transcript, are the output.
- It’s designed for retention research. Fluency vs. storage strength, desirable difficulty, and the zone of proximal development are baked into the instructions, not buzzwords.
- It’s topic-agnostic. Yoga, fitness, theory, and code all appear in the skill’s own examples.
- It refuses to teach aimlessly. No mission, no lessons — it interviews you first.
- It’s free. MIT license, and the whole thing is plain markdown and HTML you can edit.
What the Skills.sh Ranking Tells You#
459,800 installs is the raw number from the skills.sh all-time snapshot we pulled in mid-August 2026. Context helps: the other members of the family ranked above it are Grill with Docs (#4, 735K) and Setup Matt Pocock’s Skills (#9, 631K) — both are one-shot configuration utilities. Teach is the only one of the three built for sustained use over weeks, so its install count likely understates how much real session time it gets. Install counts on skills.sh measure adds, not usage.
Common Mistakes to Avoid#
Mistake #1: Running it in your main repo. Teach treats the current directory as the workspace, so running it in a project repository mixes lessons with source code. A dedicated folder per topic keeps MISSION.md and lessons out of commits.
Mistake #2: Skipping the mission to save time. The skill interrogates you about why you want to learn for a reason — every later decision on what to teach traces back to it. Pushing through with a vague goal produces the abstract lessons the spec explicitly warns against.
Mistake #3: Expecting one lesson to cover a topic. Lessons are intentionally small, one tangible win each, sized for working memory. A single session may produce several lesson files, each building on the last.
Mistake #4: Deleting old learning records. Corrected misconceptions are marked ‘superseded by LR-NNNN’ instead of deleted, because the history of how your understanding shifted is itself signal. Trashing the older records erases that trajectory.
Start With One Mission#
Concrete beats abstract, per the format’s own rule. Before you run /teach, write down the real-world outcome — “ship a Rust CLI to my team” beats “learn Rust” — and check whether it fits on one screen. Then run npx skills@latest add mattpocock/skills, pick teach, and start a fresh directory for the topic. The Teach skill intro walks through the workspace layout in detail, and if the family’s engineering skills are also on your list, the Setup Matt Pocock’s Skills guide covers the one-time config pass they all share.
Related Articles
Copilot vs Cursor vs Claude Code: Best AI Coding Agent 2026?
GitHub Copilot vs Cursor vs Claude Code on 2026 pricing, agent power, model access, and terminal-vs-IDE fit, plus a practical pick for your stack.
How to Install AI Agent Skills in Claude Code and Cursor
Install AI agent skills in Claude Code, Cursor, and OpenCode using the skills.sh CLI. Step-by-step, from npm setup to running your first skill.
FAQ: Installing AI Agent Skills in Claude Code and Cursor
Answers to the most common questions about installing AI agent skills — from skills.sh CLI to Claude Code and Cursor setup, prerequisites, and troubleshooting.
Best AI Coding Assistants in 2026: A No-Nonsense Roundup
Seven AI coding assistants tested on real projects. Comparison table included. Claude Code, Cursor, Copilot, Codex CLI, Windsurf, Continue, and Amazon Q — which one actually saves time.