Jane Street Open-Sources a Socratic Tutor That Makes AI Stop Lecturing
Jane Street open-sourced a Socratic tutor skill that drills LLMs into question-driven teachers with diagnostics, level maps, and spaced review.
- Jane Street published its first public AI skill, a Socratic tutor, on GitHub under MIT license.
- The skill uses diagnostic placement, a level 0-5 skill map, one question at a time, spaced review, and a final exam.
- Triggers only on phrases like "teach me" or "quiz me", refusing one-off "explain X" questions.
- Ships as a single SKILL.md file, no code, portable to Claude Code, ChatGPT Projects, or any system prompt.
- Repo crossed 1,600 stars within days, signaling strong demand for reusable agent skill packages.
- Notable as a rare open-source release from a secretive quant firm known for internal AI tooling.
Jane Street has published a small open-source package that turns a compatible large language model into a structured personal tutor. The MIT-licensed agent_skills repository contains one skill, socratic-tutor, and has surpassed 1,600 GitHub stars.
The release consists of a directory and a SKILL.md instruction file. That file directs the model to teach primarily by asking questions that force the user to reason, predict, and explain, not by lecturing.
Its compact design shows how an agent can acquire a specialized workflow through written instructions alone.
A tutor that makes the learner work
The instructions require the host model to ask focused questions, wait for each response, test understanding, and limit automatic praise. The skill declares triggers such as teach me
, tutor me
, quiz me
, drill me
, and help me master
. Requests such as explain X
or what is X
receive a direct answer and skip the tutoring session.
An intent guard controls entry into the workflow. Clear tutoring requests start a diagnostic, isolated questions follow the assistant’s normal behavior, and ambiguous requests prompt one brief clarifying question. This guard prevents the model from turning a simple query into an unsolicited quiz.
Inside the question loop
After activation, the model follows a defined sequence from initial placement to final assessment:
| Stage | Model instruction | Purpose |
|---|---|---|
| Diagnostic | Ask questions that reveal the learner’s current understanding. | Start at an appropriate level. |
This story is for Pro members
You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.