Jane Street Open-Sources AI Socratic Tutor Skill
Original: Jane Street: Socratic Tutor
Why This Matters
Shows how structured prompt specs can encode pedagogy into reusable AI agent behaviors.
Jane Street published a Socratic tutoring agent skill on GitHub, designed to teach users through guided questioning rather than lecturing, covering diagnostics, skill maps, and spaced review sessions.
Jane Street's open-source `agent_skills` repo now includes a Socratic Tutor skill—a structured prompt spec for AI agents that transforms passive Q&A into active, Socratic-style learning sessions. The skill activates on specific trigger phrases like 'teach me', 'tutor me', 'quiz me', or 'drill me', and explicitly excludes one-off explanation requests ('what is X', 'explain X'), which the agent handles normally.
The workflow is methodical. Before anything else, the agent confirms the user actually wants a tutoring session—a guard against false starts. It then breaks the topic into 4–10 concrete subskills (or a two-level tree for broad subjects), runs a 3–5 question diagnostic spanning easy to hard, and only then builds a skill map rating the user from levels 0–5 per subskill. Teaching happens through questions that force the user to reason, predict, and explain—not through lectures. The session closes with spaced review and a final assessment.
The spec is 102 lines of Markdown, authored as a reusable skill definition rather than a standalone app.