MIT-licensed agent skill from Jane Street that turns a model into a strict Socratic tutor. It places you with a short diagnostic, tracks a skill map with levels 0-5, asks one question at a time, and mixes in spaced review and a final assessment.
The skill only starts when you ask to be taught, quizzed or drilled. For a one-off "explain X" it answers normally, and when intent is unclear it asks one short question first. A session breaks the topic into 4-10 subskills, runs 3-5 diagnostic questions before teaching anything, and renders an ASCII skill map after every level change.
The rules are strict on purpose. One question per message, nothing after it. A wrong answer gets a narrower question that exposes the contradiction, not the answer, and a direct explanation only after 2-3 failed attempts. A level goes up only after 2+ independent checks, one of them unprompted recall, and earlier subskills get re-tested without warning.
Install it with npx skills add https://github.com/janestreet/agent_skills --skill socratic-tutor, copy the folder into your agent's skills directory, or paste SKILL.md into a chat as a system prompt.