Jane Street Agent Skills Release: 5,794 Stars in 8 Days
The janestreet organization on GitHub, the account of quantitative trading firm Jane Street, has published agent_skills, a collection of agent skills released under the MIT license. The repository was created on November 3, 2026 and reached 5,794 stars and 401 forks in about eight days, an average of more than 700 stars a day. The striking part is the size of the collection: it contains exactly one skill.
That mismatch between attention and content is the story. Skill collections have mostly come from individual developers and tool makers. A large financial firm publishing in the same shared format, and drawing this much attention for a single tutoring skill, suggests the SKILL.md format is becoming a common packaging layer for agent behavior, one that organizations outside the tooling industry now treat as worth publishing in.
A Collection With One Skill In It
Each skill in the repository is a directory containing a SKILL.md file that holds instructions for an agent. The format is deliberately plain: no SDK, no runtime, just a structured document an agent reads and follows. The only published entry so far is socratic-tutor, which turns an LLM into a strict private tutor.
The star count needs a careful reading. Stars measure curiosity, not use, and a famous firm publishing an unusual artifact is itself newsworthy, so part of the traffic is attention to Jane Street rather than to tutoring. Still, 401 forks point in a more practical direction. People fork a skill repository to copy and adapt its contents, and a fork count that high within eight days means a meaningful share of visitors came to take the format, not just to look at it.
How socratic-tutor Teaches
The skill runs a structured teaching loop. It starts by diagnosing the learner, then maps progress on a scale from stage 0 to stage 5. Teaching proceeds one question at a time, in the Socratic style the name promises, rather than through lectures. Material comes back through spaced review, and the process ends with a final assessment.
As a design, this is a reasonable encoding of established tutoring practice: diagnosis before content, small steps, retrieval practice through review, and a closing check. What a SKILL.md cannot guarantee is follow-through. The instructions shape how a model behaves in a session, but pacing across days, honest scoring of the final assessment, and resistance to a learner who simply asks for the answer all depend on the model running the skill. No third-party evaluation of the tutor's actual learning effect has appeared, so the design is documented while the outcome is not.
Two Ways To Use A Skill
| Setup | How it works |
|---|---|
| Coding agent with SKILL.md support | Copy the skill directory into the agent's skills folder |
| Plain chatbot | Paste the SKILL.md content into the conversation, or into a custom or system prompt |
The second path is the more telling one. A skill that works by being pasted into any chatbot is portable across models and vendors, which is exactly why a plain text format can spread faster than a plugin system tied to one platform. The trade-off is control: a pasted prompt carries no versioning, no permissions, and no guarantee that the host model will honor every rule in the file. For a tutor that trade-off is harmless. For skills that touch code, files, or money, it becomes the central design question, and it is one this repository does not have to answer yet.
Why A Trading Firm Publishing Skills Matters
Jane Street is known for quantitative trading, not for developer education products, and nothing in the repository explains the firm's purpose. It may be recruiting signal, internal tooling shared on the way out, or an experiment. The purpose is unconfirmed, and so is any internal use.
The format choice is the part that can be read with confidence. The firm could have published a blog post about tutoring with LLMs, or a standalone app. It chose a SKILL.md directory that coding agents can install directly, which means it expects its audience to be running agents that consume this format. When a finance firm treats agent skills as a publication channel, the skill ecosystem stops being a niche of agent tooling and starts looking like a general distribution format for procedural knowledge, closer to publishing a checklist than shipping software.
What Is Not Known Yet
The repository currently lists one skill. How many skills Jane Street plans to add, why the collection was published, whether it is used inside the firm, and whether socratic-tutor improves learning outcomes are all unconfirmed as of November 12, 2026.
- Collection size: one published skill, with no announced roadmap for more.
- Purpose: no stated reason for the release, and no confirmation of internal use.
- Effectiveness: no third-party verification of the tutor's learning effect.
None of those gaps reduce the signal in the numbers. A one-skill repository from a trading firm gathered 5,794 stars in eight days because the format is legible to a large and growing audience of agent users. Whether agent_skills grows into a real library or stays a single well-made example, it marks the point where agent skills started arriving from outside the agent tooling world.
FAQ
What is janestreet/agent_skills?
It is a GitHub repository from the janestreet organization, created on November 3, 2026 and released under the MIT license. Each skill is a directory containing a SKILL.md file with instructions for an agent. As of November 12, 2026, the repository contains one published skill.
What does the socratic-tutor skill do?
socratic-tutor turns an LLM into a strict private tutor. It diagnoses the learner, maps progress across stages 0 to 5, asks one question at a time, uses spaced review, and finishes with a final assessment.
How do you install a skill from agent_skills?
With a coding agent that supports the shared SKILL.md format, copy the skill directory into the agent's skills folder. With a plain chatbot, paste the SKILL.md content into the conversation or into a custom or system prompt.
How fast did agent_skills grow on GitHub?
The repository reached 5,794 stars and 401 forks in about eight days after its November 3, 2026 creation, an average of more than 700 stars a day, as observed on November 12, 2026.
Will Jane Street publish more skills in this repository?
That has not been confirmed. The current list contains only socratic-tutor, and Jane Street has not stated a purpose for the collection, whether it is used internally, or how many skills it may grow to.
Related Reading
- Hallmark v2 Design Skill: 50 Themes for Coding Agents, another SKILL.md package that spread through coding agents
- day-trader-skill: 11 Claude Skills, Time Machine Replays, a larger skill collection built around a single workflow