One teacher. One mind.
Portable agent skills that run Eero Alvar’s AI learning loop.
npx skills add vasanthsreeram/Alvarmethod -g --allThat’s the install. Works in Claude Code, Codex, Grok, Pi, OpenCode, Cursor, and the rest of the skills CLI agents.
Then open a learning folder (not this repo) and say:
/teach I want a solid introduction to <topic>
You should get a quiz picker (not A/B/C/D in chat) → a mermaid DAG → one reasoning step → a lock-in quiz.
The open skills CLI. Discovers every skills/*/SKILL.md in this repo.
# global, every detected agent, no prompts
npx skills add vasanthsreeram/Alvarmethod -g --all
# list what’s in the pack
npx skills add vasanthsreeram/Alvarmethod --list
# one skill
npx skills add vasanthsreeram/Alvarmethod --skill teach -g -y
# one harness
npx skills add vasanthsreeram/Alvarmethod -g -y -a claude-code -a grok -a pi -a opencode -a codexUpdate later:
npx skills update -gNo extra CLI — runs install.sh from GitHub:
npx github:vasanthsreeram/Alvarmethod
npx github:vasanthsreeram/Alvarmethod --list
npx github:vasanthsreeram/Alvarmethod --claude --grokgit clone https://github.com/vasanthsreeram/Alvarmethod.git
cd Alvarmethod
./install.sh # all known agent dirs
./install.sh --project
./install.sh --uninstall| Agent | Global path |
|---|---|
| Codex | ~/.codex/skills + ~/.agents/skills |
| Claude Code | ~/.claude/skills |
| Grok | ~/.grok/skills |
| Pi | ~/.pi/agent/skills |
| OpenCode | ~/.config/opencode/skills |
Credit: Eero Alvar — How I Use AI to Learn Things (2026-08-14). This pack implements that loop as Agent Skills. It is not his original Pi quiz / Obsidian setup. Captions: source/eero-alvar-how-i-use-ai-to-learn-things.md.
Many-to-many learning leaks:
| Direction | Waste |
|---|---|
| One outlet → many students | Never fitted to your edge |
| One student → many outlets | Switching cost + the brain hedges on sources it does not trust |
Fix: one interface, many sources. Trust is engineered (verify). Struggle stays in the material. The system eats logistics.
flowchart LR
A[Probe<br/>edge of this mind] --> B[Plan<br/>mermaid DAG]
B --> C[Teach<br/>one node]
C --> D[Lock-in quiz]
D -->|pass| C
D -->|fail| E[Insert prereq]
E --> C
- Probe — graded MCQ in the harness picker, start wide, binary-search every strand
- Plan — mermaid DAG for this mind, shown before teaching
- Teach — one reasoning step, picture if needed, quiz, next node
| Skill | Job |
|---|---|
teach |
Full loop: probe → plan → one-step teaching |
probe |
Understanding map only |
learn-profile |
Write .alvar/LEARNER.md |
learn-visual |
One SVG, look at it, fix it |
learn-verify |
Fact-check a claim before it is taught as fact |
Quizzes must use the native picker:
| Agent | Tool |
|---|---|
| Grok / Codex | ask_user_question |
| Claude Code | AskUserQuestion |
| OpenCode | question |
| Pi | quiz / ask_user / askUserQuestion |
If the agent pastes A/B/C/D, say “use the quiz tool.” If it dumps a textbook, say “one node only.”
Session files land in the learning folder:
.alvar/LEARNER.md
.alvar/maps/<topic>.md
.alvar/sessions/<date>-<topic>.md
.alvar/visuals/…
Point Obsidian at .alvar/ if you want LaTeX preview.
See CONTRIBUTING.md. Edit skills here, then reinstall. Don’t patch the copies under ~/.claude / ~/.grok.
MIT for the skills, installer, and docs.
The video and spoken method are Eero Alvar’s. Watch the original: youtu.be/kzcI5F4tGiU.