Alvarmethod skill
Five skills turn Claude Code or Codex into a tutor: a knowledge probe, a topic map, one step of teaching and a lock-in quiz.
- 167 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- Teaching instructions only; the skills create `.alvar/` files in the learning folder and draw SVGs, and run nothing else.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/alvarmethodDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- Teaching instructions only; the skills create `.alvar/` files in the learning folder and draw SVGs, and run nothing else.
This is third-party code. Review the repository files before installing.
What it does
A set of five skills implements the probe, plan and teach learning loop from Eero Alvar's video "How I Use AI to Learn Things". The command `/teach I want a solid introduction to <topic>` starts the `teach` skill: `probe` first asks multiple-choice questions through the agent's own picker and finds the edge of your knowledge, then a mermaid topic graph is built and shown before any teaching. The agent then explains one reasoning step and locks it in with a quiz; a pass moves to the next node, a fail inserts the missing prerequisite. The other skills handle supporting jobs: `learn-profile` keeps a learner profile, `learn-visual` draws one SVG and checks it, `learn-verify` fact-checks a claim before it is taught as fact. Session files (`.alvar/LEARNER.md`, topic maps, lesson logs) sit in the learning folder. The author notes this is an implementation of the method; Eero Alvar's original Pi and Obsidian setup is different.
Who it is for. People learning a new topic with Claude Code or Codex who want step-by-step explanations at their level instead of a generic summary.
Good fit when
- You are picking up a new field such as databases or cryptography and do not know where to start.
- The agent produces an encyclopedic wall of text and you want one step plus a check that it landed.
- Before an exam you want to find gaps with a short series of multiple-choice questions.
- An explanation needs a diagram that the agent draws in SVG and checks itself.
- An important claim has to be verified before you accept it as fact.
Not a fit when
- You need a quick answer to a specific question: the loop with a topic map and quizzes will slow you down.
- The agent has no built-in choice picker: the skill requires it for quizzes and does not print A, B, C, D in chat.
- Two people learn together: the profile files assume a single learner.
Example request
/teach I want a solid introduction to how databases work: first probe what I already know, show the plan and teach one step at a time.Limitations
The skills assume an agent with its own choice tool: `AskUserQuestion` in Claude Code, `ask_user_question` in Codex and Grok, `question` in OpenCode. If the agent pastes options as letters into chat, repeat the request as "use the quiz tool". Run `/teach` in a separate learning folder, not in this repository. The skills create `.alvar/` files with the profile and topic maps, which you keep yourself. The project implements the method from the video and does not serve as the method author's original setup.
How to disable. Run `npx skills remove` for this set or delete the `teach`, `probe`, `learn-profile`, `learn-visual` and `learn-verify` folders from the agent's skills directory. Delete the `.alvar/` files in learning folders by hand.
FAQ
How do I install Alvarmethod in Claude Code?
Run `npx skills add vasanthsreeram/Alvarmethod -g --all`: the skills install into every detected agent. One skill installs like this: `npx skills add vasanthsreeram/Alvarmethod --skill teach -g -y`. A variant without the skills CLI is `npx github:vasanthsreeram/Alvarmethod`.
How do I start a lesson?
Open a session in a separate learning folder and type `/teach I want a solid introduction to <topic>`. A choice picker appears first, then a mermaid topic graph, then the first step of teaching.
Why does the agent print options A, B, C, D in chat?
It skipped the built-in choice tool. Say "use the quiz tool" and repeat: in Claude Code that is `AskUserQuestion`. If the agent dumps a textbook, ask for "one node only".
Which skills are in the set?
Five: `teach` (the whole loop), `probe` (understanding map), `learn-profile` (learner profile), `learn-visual` (one SVG with a check) and `learn-verify` (claim check before teaching).
Where is lesson data stored?
In the learning folder, under `.alvar/`: `LEARNER.md` with the profile, `maps/` with topic maps, `sessions/` with lesson logs and `visuals/` with drawings. Open `.alvar/` in Obsidian for LaTeX preview.
Discussion
No questions or reviews yet.
Related
SkillsA mentor skill for Claude Code that guides you to LeetCode solutions with hints instead of handing over code at once
Plugin and skill for Claude Code, Codex and Kilo that researches a topic on the web and builds a local learning board
A personal AI tutor for Claude Code: a syllabus, one lesson at a time, and each next lesson shaped by your annotations and feedback
A tutor for any subject: diagnosis, a study plan, project tasks and mastery checks, with the results saved to Markdown files
A tutoring skill for STEM and AI: it finds the knowledge gap first, then teaches one step and checks understanding
Claude Code plugin that builds a study plan for any topic, runs adaptive quizzes and schedules reviews with the SM-2 algorithm.
README in short
The README gives three install routes and explains the method's idea: one interface instead of many sources, with trust built by verification. After that come a diagram of the loop, a table of skills and a list of choice tools per agent.
README badge
Original SKILL.md
--- name: teach description: > One-to-one tutor using the Alvar method — probe the edge of this mind, plan a mermaid DAG, then teach one reasoning step at a time with a lock-in quiz. Use when the user runs /teach or /skill:teach, or says teach me, tutor me, I want to learn, introduce me to, walk me through, or asks for a lesson fitted to what they already know. license: MIT metadata: author: vasanthsreeram version: "1.0" source: "Eero Alvar — How I Use AI to Learn Things https://youtu.be/kzcI5F4tGiU" --- # Teach (Alvar method) You are **one teacher for one mind**. Not a course. Not a survey. Read, in order, before the first question: 1. [references/philosophy.md](references/philosophy.md) 2. [references/process.md](references/process.md) 3. [references/learner-files.md](references/learner-files.md) 4. [references/quiz-ui.md](references/quiz-ui.md) — **required.** Quizzes use the harness question tool. Never A/B/C/D in chat. 5. `.alvar/LEARNER.md` if it exists (else offer `learn-profile` or write a stub from [assets/LEARNER.md](assets/LEARNER.md) after 3–5 questions) If `probe`, `learn-visual`, or `learn-verify` are installed, use them for those jobs. If not, follow the same protocols inline. ## Hard rules - Struggle stays in the material. You absorb logistics. - Do not reteach `known`. Do not start in `unknown` with no ramp. - One reasoning step per turn. Stop. Quiz that step. Advance only on lock-in. - Show the mermaid plan **before** teaching. Do not skip the graph. - Never dump the whole explanation in one message. - Do not invent citations. Verify or mark uncertainty. - Trust is engineered: if a claim matters and you are not sure, verify before teaching it as fact. ## Turn 0 1. Restate the goal in one sentence. Confirm. 2. Load or create learner files. 3. Probe (or reuse a map for this exact goal if it is fresh and the user agrees). 4. Plan + mermaid. Pause. 5. Teach node 1. ## Quiz shape Call the harness quiz UI in [references/quiz-ui.md](references/quiz-ui.md). Do not paste letters in chat.