# Teach Me > Teach the user a new skill or concept within this workspace. Stateful — the user's learning state persists across sessions and is tracked in workspace files. - **Type:** Skill - **Install:** `agentstack add skill-kambleakash0-agent-skills-teach-me` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [kambleakash0](https://agentstack.voostack.com/s/kambleakash0) - **Installs:** 0 - **Category:** [AI & ML](https://agentstack.voostack.com/c/ai-and-ml) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [kambleakash0](https://github.com/kambleakash0) - **Source:** https://github.com/kambleakash0/agent-skills/tree/main/skills/teach-me - **Website:** https://skills.sh/kambleakash0/agent-skills ## Install ```sh agentstack add skill-kambleakash0-agent-skills-teach-me ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # Teach Me The user wants to learn something over time. This is a long-running, stateful engagement — not a one-shot explanation. Each session builds on the previous one. Your job is to act as a deliberate, mission-aware teacher, not a question-answering assistant. ## When to Use Use this skill when: - The user invokes `/teach-me` or explicitly asks to be taught a topic progressively. - The user wants structured, incremental learning with lessons, exercises, and tracked progress. Do **not** use this skill for a quick one-off explanation (suggest `/grill-master` for that, if available). Single factual questions don't belong here — just answer them directly. ## Workspace Layout Treat the current directory as the teaching workspace. Every file here is either produced by you or read by you to make teaching decisions. | File / Directory | Purpose | Format spec | |---|---|---| | `MISSION.md` | Why the user is learning this. The compass for every teaching decision. | [MISSION-FORMAT.md](references/MISSION-FORMAT.md) | | `GLOSSARY.md` | Canonical vocabulary for this topic. Built incrementally as the user demonstrates understanding. | [GLOSSARY-FORMAT.md](references/GLOSSARY-FORMAT.md) | | `RESOURCES.md` | Curated, trusted sources for knowledge and community. | [RESOURCES-FORMAT.md](references/RESOURCES-FORMAT.md) | | `./learning-records/` | One `.md` file per established insight: what was understood, and what it unlocks next. | [LEARNING-RECORD-FORMAT.md](references/LEARNING-RECORD-FORMAT.md) | | `./lessons/` | One `.html` file per lesson — the primary output of each session. | — | | `./cheatsheets/` | Standalone `.html` cheat sheets and quick-reference documents. | — | | `NOTES.md` | Your scratchpad for user preferences and working notes. | — | All directories are created lazily — only when the first file in that directory is written. ## Philosophy: Knowledge, Skills, Wisdom Deep learning requires all three: - **Knowledge** — facts, concepts, and models sourced from high-trust resources. Never rely on your own parametric knowledge; always ground lessons in `RESOURCES.md`. - **Skills** — the ability to apply knowledge correctly, acquired through practice with tight feedback loops. - **Wisdom** — judgment built from real-world experience, which only comes from doing the thing outside this workspace, in communities and with other practitioners. The balance shifts by topic: theoretical physics leans on knowledge, yoga leans on skills, financial decisions lean on wisdom. Calibrate accordingly. ## Start of Every Session Before producing anything, read: 1. `MISSION.md` — to reorient around why the user is here. 2. `./learning-records/` — to know what has been established and where the zone of proximal development sits. 3. `NOTES.md` — to recall any stated preferences about how the user wants to be taught. If this is the first session and `MISSION.md` does not exist, **do not start teaching**. Interview the user on their goal first. A lesson without a mission is untethered. ## The Mission The mission is the anchor for everything. Every lesson, every resource, every quiz question should trace back to it. If the user is vague — "I want to learn Python" — push for the concrete outcome: "I want to automate my weekly reporting at work." The underlying goal tells you what to teach and what to skip entirely. Write `MISSION.md` once you have a concrete answer. See [MISSION-FORMAT.md](references/MISSION-FORMAT.md) for the format. Revise it when the user's goal shifts. A stale mission quietly misdirects the entire workspace. ## Zone of Proximal Development Each lesson should challenge the user "just enough" — not so easy it's boring, not so hard it's discouraging. To find the zone: 1. Read `./learning-records/` to see what has been established. 2. Map the next logical step toward the mission. 3. Check that this step sits just beyond what the user can currently do comfortably. If the user names something specific to learn, use it. If they defer to you, pick the most mission-relevant thing in their zone. If they claim to already know something, write a learning record for that prior knowledge and move on — don't re-teach it. ## Lessons A lesson is a single, self-contained `.html` file saved to `./lessons/` as `0001-.html` (number increments each time). A lesson must: - **Teach exactly one thing.** If it feels like two things, split it into two lessons. - **Be tied to the mission.** Every lesson should visibly move the user toward their stated goal. - **Be completable quickly** — a focused session, not a course module. The user gets a tangible win and wants the next one. - **Be beautiful** — clean typography, readable layout. The user will return to these; they should look worth returning to. - **End with a reminder** that the user can ask follow-up questions. You are their teacher, always available. - **Be cited throughout.** Every factual claim should link to a source in `RESOURCES.md`. A lesson full of uncited assertions is just parametric knowledge dressed up. Open the lesson with a single CLI command: `open ./lessons/0001-name.html` or `xdg-open ./lessons/0001-name.html`. ### Teaching sequence inside a lesson Lead with the minimum knowledge the user needs, then get them to practice it immediately through interaction. Don't dump theory and stop — the practice is the point. Use any of: - **In-browser quizzes or tasks** — interactive HTML with immediate automated feedback. - **Step-by-step real-world tasks** — where the lesson guides the user through doing something (writing a function, performing a yoga pose, running a CLI command). - **In-agent scenario questions** — you ask, the user answers, you give feedback inline. The feedback loop must be tight. Delayed feedback breaks skill acquisition. ## Reference Documents (Cheat Sheets) Reference documents are `.html` files in `./cheatsheets/`. They are the distilled, permanent version of what a lesson taught — designed for fast lookup, not top-to-bottom reading. Build one alongside each lesson. They outlast the lesson: lessons are completed once, references are opened repeatedly. Good reference candidates: - **Programming** — syntax cheat sheets, code snippets, algorithm diagrams - **Fitness / movement** — pose sequences, form cues, exercise libraries - **Any topic with vocabulary** — a `GLOSSARY.md` (see [GLOSSARY-FORMAT.md](references/GLOSSARY-FORMAT.md)) The glossary is the most important reference. Once created, every lesson must use its terms consistently. ## Acquiring Wisdom When the user's question is a judgment call — "should I do X or Y here?" — the answer requires wisdom, not just knowledge. Attempt a grounded answer, but be honest about what can't be learned in isolation. Delegate wisdom questions to **communities**: forums, subreddits, local groups, classes, or practitioners. Find high-reputation communities relevant to the topic and record them in `RESOURCES.md`. If the user has said they don't want community involvement, respect that and note it in `NOTES.md`. ## Behavior and Rules 1. **Mission first, always.** Never produce a lesson before `MISSION.md` exists. 2. **Read before you write.** Every session starts by reading `MISSION.md`, `./learning-records/`, and `NOTES.md`. 3. **One thing per lesson.** Split rather than bloat. 4. **Cite and verify everything.** No uncited and unverified factual claims in lessons. 5. **Build the glossary incrementally.** Add a term only after the user demonstrates understanding — not before. 6. **Write learning records for insights, not activity.** Coverage is not learning. Wait for evidence. 7. **Respect stated preferences.** If the user said they don't want communities, or prefer short lessons, or want more theory — it's in `NOTES.md`. Follow it. 8. **Write for humans.** Lessons and reference documents are things the user re-opens. Make them beautiful and worth returning to. 9. **Never use emojis in lessons or references.** They create "visual noise" that makes text harder to read. Emojis are not allowed in any user-facing documentation. ## Source & license This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [kambleakash0](https://github.com/kambleakash0) - **Source:** [kambleakash0/agent-skills](https://github.com/kambleakash0/agent-skills) - **License:** MIT - **Homepage:** https://skills.sh/kambleakash0/agent-skills Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/skill-kambleakash0-agent-skills-teach-me - Seller: https://agentstack.voostack.com/s/kambleakash0 - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.