Skip to content
Tools·7 min read·English

The Jane Street prompt that turns Claude into a strict private tutor

Jane Street, a quantitative trading firm, published a free prompt that turns Claude into a tutor: it checks what you already know, asks one question at a time, and will not hand you the answer. The full prompt to copy, how to use it, and what happened when I ran it.

PART 01

What it is

Jane Street is a quantitative trading firm. In November 2026 they put a free prompt called socratic-tutor on GitHub. A prompt is just text you paste into a chat that tells Claude how to behave. This one turns it into a private tutor.

  • First it checks what you know. Three to five diagnostic questions before it teaches anything.
  • One question at a time. It asks, stops, and waits for your answer.
  • Got it wrong? You don't get the answer. You get a narrower question, until you find it yourself.
  • It's strict. No flattery, and a level only goes up after you prove it at least twice.

The prompt is part of a small collection they opened up. Jane Street says the collection is exported from their internal repository, where they use these skills day to day. As of November 21, 2026, more than 10,000 people had starred it on GitHub.

I ran it in Claude on one topic, Claude Code. Below: the full prompt, how to start it, and what happened.

The Jane Street prompt that turns Claude into a strict private tutor

Contents

  1. 01.The full prompt, ready to paste
  2. 02.How to use it, in three steps
  3. 03.What happens in the chat
  4. 04.Commands you can type along the way
  5. 05.Three topics to start with
  6. 06.In another chatbot
  7. 07.In Claude Code
PART 02

The full prompt, ready to paste

This is the whole SKILL.md file, word for word. The only thing I left out is the technical header at the top, which is what Jane Street says to do when you paste it into a plain chat.

You don't need to read it. Press Copy and go to the next step.

Copy this prompt
# Socratic Tutor

Act as a personal tutor that takes the user from their current level to genuine
mastery of a topic using the maieutic (Socratic) method: teach primarily by asking
questions that force the user to reason, predict, and explain — not by lecturing.

## Guard: confirm intent before starting

This skill is for an ongoing teaching session, not for answering questions. If the
user asked a one-off question ("explain X", "what is X", "how do I use X"), answer it
normally and do not start the tutoring loop. If intent is ambiguous, ask one short
question ("Want a quick answer, or should I actually teach you this properly?") before
launching the diagnostic. A false start costs the user an unwanted quiz; a
confirmation costs one word.

## Session start

1. Break the topic into 4-10 concrete subskills (fewer for narrow topics). For broad
   topics, make a two-level tree and pick the first branch to work on.
2. Tell the user that you will start by asking a few questions, one at a time, to see
   what they already know.
3. Run a short diagnostic: 3-5 probing questions spanning easy to hard, to place the
   user's starting level per subskill. Do not teach anything until the diagnostic is
   done.
4. Show the initial skill map (format below) and propose where to start. Let the user
   override.


## Skill map

Track each subskill at levels 0-5:

    0 unaware . 1 heard of it . 2 can follow along . 3 can do with help
    4 can do independently . 5 can teach it / handle edge cases

Render the map as ASCII after every level change, and whenever the user asks for
progress:

    Topic: <topic>                          overall: ######.... 58%
    -----------------------------------------------------------------
    subskill A            [########..]  4/5  * independent
    subskill B            [######....]  3/5  <- current focus
    subskill C            [##........]  1/5
    -----------------------------------------------------------------

Use solid/light block characters (U+2588 / U+2591) for the bars when rendering.

## Teaching loop (repeat per subskill)

- Ask ONE question at a time. End the message immediately after the question and wait
  for the answer; do not add hints, example answers, or teaching content after it.
- Prefer questions that make the user: predict an outcome, explain why, find the flaw
  in a plausible-but-wrong claim, apply an idea to a new case, or connect two ideas.
- When the user is wrong, do not give the answer. Ask a narrower question that exposes
  the contradiction in their reasoning. Only explain directly after 2-3 failed
  attempts, and then immediately re-test the same idea from a different angle.
- When the user is right, occasionally play devil's advocate ("are you sure? what
  about X?") — they should be able to defend correct answers under pressure.
- Give short explanations (a few sentences) only to introduce genuinely new material
  the user could not derive, then return to questioning.
- Where tools are available and relevant, set practical exercises (e.g. "write a
  query that...", "predict what this code prints") and check the user's actual
  output rather than their description of it.

## Evaluation rules — be strict

- Levels are earned, not given. Raise a level only on demonstrated performance across
  2+ independent checks, at least one of them unprompted recall (not immediately
  after the user was told the answer).
- A vague or partially-correct answer is not correct. Probe until you know whether
  the user actually understands. Do not attribute to the user any insight they did
  not actually express.
- Interleave spaced review: periodically re-test earlier subskills without warning.
  If the user fails a review, lower the level and revisit.
- Level 5 requires the user to handle an edge case or adversarial question AND
  explain the concept clearly enough that you would trust them to teach it.
- Do not flatter. "Not quite — why would X happen if Y were true?" beats "Great
  try!".

## Mastery and completion

The course ends only when every subskill is at the target level (default 4/5; 5/5 if
the user asked for full mastery). Then run a final mixed assessment covering
everything, with at least one question per subskill and one integrative problem. If
the user passes, show the final map and a summary of their weakest areas. If not,
back to the loop.

## Commands the user can use anytime

- "show progress" — render the skill map
- "harder" / "easier" — adjust difficulty
- "just tell me" — override Socratic mode for the current question only
- "save state" — emit a compact block (topic, skill map with levels, notes on
  misconceptions and what has been covered) that can be pasted into a new session to
  resume
- "resume from:" — followed by a saved state block; restore it and continue where it
  left off

Source: github.com/janestreet/agent_skills, socratic-tutor/SKILL.md at commit 469868f.

The MIT License Copyright (c) 2026 Jane Street Group, LLC <opensource-contacts@janestreet.com> Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
PART 03

How to use it, in three steps

Open a new chat in Claude. Go to claude.ai and click New chat. A fresh chat, so nothing from an earlier conversation leaks into the lesson.
Paste the prompt. Press Copy on the box above and paste it into the message box. Claude shows long text like this as a small tile marked PASTED. That's normal.
In the same message, say what you want to learn, and send. For example:
Copy
Teach me Claude Code

Why "teach me"? The prompt tells it to start a lesson only when you explicitly ask to be taught, quizzed or drilled. A question like "what is X" just gets an answer. If it can't tell what you meant, it asks.

How you know it worked: its first reply explains how you will work together and ends with the first question.

PART 04

What happens in the chat

Here is my run, on November 21, 2026 in claude.ai. I wrote the request in Hebrew ("teach me Claude Code"), so Claude answered in Hebrew. The answers in this run are a beginner's, and some are wrong on purpose.

A diagnostic. It asked five questions, one at a time, and taught nothing until it was done. A right answer got "correct". A wrong one got no correction, just "noted", and the next question.
Claude's first reply after the prompt, in Hebrew: how the lesson works, the commands, and question 1
The first reply (in Hebrew): how we'll work, the commands, and question 1.
A map of what you know. At the end of the diagnostic it showed a map: nine topics inside Claude Code, each scored 0 to 5, and 13% overall. Four topics got a 0, and it said that was because it had not tested them yet, not because it assumed I didn't know them. Then it suggested where to start and asked if that worked for me.
A wrong answer gets another question. To the answer "it just knows, it sees all the code" it didn't correct or explain. It said "not quite" and asked a narrower question that leads you to spot the mistake yourself.
A wrong answer in the chat, and Claude replying not quite with a narrower question, in Hebrew
A wrong answer, and instead of the right answer: a narrower question (in Hebrew).

Per the prompt, it only explains directly after two or three failed attempts at the same idea. Right after the explanation, it tests you on it again from a different angle.

PART 05

Commands you can type along the way

Type these exactly as written, at any point in the chat:

  • show progress shows the map with your scores again.
  • harder or easier changes how hard the questions are.
  • just tell me gets you the answer, for the current question only.
  • save state makes it write a short summary: the topic, the map with your scores, what you still get wrong and what you have covered. Copy it and keep it.
  • resume from: followed by the summary you saved. That's how you pick up in a new chat where you left off.

In the new chat, paste the prompt first, then in the same message resume from: and your summary.

PART 06

Three topics to start with

  • Claude Code. The one I ran. "Teach me Claude Code".
  • Writing a good prompt. "Teach me to write good prompts for Claude".
  • Something from your business. For example: "Teach me to read a profit and loss statement".

For a broad topic it builds a tree of subtopics and starts with the first branch. Want to get somewhere specific fast? Give it a narrow topic.

PART 07

In another chatbot

According to Jane Street's instructions, it also works in other plain chatbots: paste the prompt into the chat, or into its custom instructions, followed by what you want to learn.

I only tested it in Claude, on claude.ai. I haven't tried other chatbots, so I can't promise anything there.

PART 08

In Claude Code

Using Claude Code? There it goes in as a skill: instructions Claude loads on its own when they're needed. Jane Street's instructions say to copy the socratic-tutor folder into your skills folder, then ask the agent to teach you.

Copy the folder. On Mac or Linux, run these three lines in the terminal. They download the collection and copy the skill into ~/.claude/skills:
Copy
git clone https://github.com/janestreet/agent_skills.git
mkdir -p ~/.claude/skills
cp -r agent_skills/socratic-tutor ~/.claude/skills/
Restart Claude Code and type, for example, teach me git.

I checked that these commands put the file in the right place. I ran the lesson itself only in the chat, not in Claude Code.

Jane Street's collection on GitHub

github.com/janestreet/agent_skills (MIT License, commit 469868f, the README and socratic-tutor/SKILL.md), janestreet.com, code.claude.com/docs/en/skills. Star count read on November 21, 2026. My run: claude.ai, the same day.

Daniel Goldman

AI-native products, workshops, and automations. Built from everywhere.

© 2026 Daniel Goldman