Most AI coding agents are brilliant at writing Python but clueless about what a real lab workflow looks like. They can pip-install Biopython, yet they cannot tell you how a medicinal chemist actually runs a docking study or how a genomicist designs a variant-calling pipeline. Scientific Agent Skills exists to fix exactly that gap: it is an open source library of 177 validated skills plus more than 100 scientific databases that teach agents like Claude Code, Cursor, and Codex the procedures of actual science, from genomics and cheminformatics to clinical research and drug discovery (GitHub, retrieved 2026-10-23).
Key Takeaways
Scientific Agent Skills (K-Dense-AI/scientific-agent-skills) is an MIT-licensed library of 177 validated skills and 100+ scientific databases for AI agents.
It is built on the open Agent Skills standard, which Anthropic donated to the Linux Foundation's Agentic AI Foundation on October 22, 2026.
The repo has roughly 54,000 stars and 4,700 forks, making it one of the most starred science-focused AI repos on GitHub.
Installation is a drop-in folder of skills that works across Claude Code, Cursor, Codex, and other agents supporting the standard.
What the project actually is
At its core, the project is a collection of procedural knowledge. Each "skill" is a folder with instructions, references, and often executable scripts that tell an AI agent not just which tool to call, but the ordered workflow a domain expert would follow. The skill set covers biology, chemistry, medicine, and drug discovery, and the library ships with access to more than 100 scientific databases (GitHub, retrieved 2026-10-23).
The project used to be called "Claude Scientific Skills." It was renamed in 2026 to reflect that it targets the open Agent Skills standard rather than a single product, and the skills now work with Cursor, Claude Code, Codex, Pi, Antigravity, and any agent supporting the standard (AgentConn, October 2026).
K-Dense, the company behind the repo, frames the idea with a sharp distinction: these skills are "the 'what to do with your tools' layer" of a scientific agent (K-Dense blog, 2026-06-02). The tools (MCP servers, APIs, databases) give the agent hands; the skills give it the lab training.
Why skills exist at all: the Agent Skills standard
A skill, per the standard, is a directory containing a SKILL.md file that starts with YAML frontmatter (a name and a description), plus optional scripts, references, and assets (agentskills.io, retrieved 2026-10-23). Skills load through progressive disclosure in three stages: Discovery (metadata in the prompt), Activation (full SKILL.md on match), and Execution (bundled files on demand). This keeps context windows small: the project's own paper reports that the always-resident descriptions of all its skills cost only 7.1% of a 200,000-token window (arXiv:2609.00065, submitted 2026-08-30).
The standard was originally developed by Anthropic and announced on October 16, 2025, then opened as a cross-platform standard in December 2025 (Anthropic engineering, published 2025-10-16). And the ecosystem just had its biggest day: on October 22, 2026, Anthropic donated the Agent Skills standard, including the name, logo, GitHub organization, and agentskills.io domain, to the Agentic AI Foundation under the Linux Foundation (Mad Robot, 2026-10-22). Anthropic counts 1 million published skills and nearly 280 million installs, with adoption by ChatGPT, Cursor, GitHub Copilot, and more than 70 agent platforms.
Aspect | Detail |
|---|---|
Project | Scientific Agent Skills |
Maintainer | K-Dense-AI (company: K-Dense) |
License | MIT |
Skills | 177 validated skills (per live repo, Oct 2026) |
Databases | 100+ scientific databases |
Compatibility | Claude Code, Cursor, Codex, Pi, Antigravity, open Agent Skills standard |
GitHub traction | ~54,000 stars, ~4,700 forks (Oct 2026) |
What 177 skills cover
The coverage spans genomics, cheminformatics, medical imaging, study design, and scientific communication, among 16 practice areas identified in the project's paper (arXiv:2609.00065, submitted 2026-08-30). The README lists domains including computational biology, drug discovery, metabolomics, proteomics, and materials science (GitHub, retrieved 2026-10-23).
Notably, the skills are validated, which matters more than the count. The AgentConn write-up notes that some skills explicitly refuse to generate synthesis routes for controlled substances, a sign the library was built with guardrails rather than just scraped together (AgentConn, October 2026).
How to use it
Installation follows the standard skill pattern: the skills are a folder of SKILL.md files plus scripts and references that drop into any agent supporting the standard. Documentation lives in the repo's `docs/` directory, with skill guides and examples (GitHub docs, retrieved 2026-10-23). K-Dense also ships K-Dense BYOK, described as a free, open source AI co-scientist desktop app with 40+ models and all 177 skills built in (GitHub, retrieved 2026-10-23).
If you work in research and already use Claude Code or Cursor, the practical move is to install the skill folder, point your agent at a real analysis task, and watch whether its workflow matches what your lab would do. That is the whole bet of the project: the difference between an agent that knows chemistry tools exist and one that knows how chemists work.
Want the 30-second version? Watch the repo signal for this project on repomaniac.com, and read this article in Indonesian for the translated edition.
Frequently Asked Questions
Is Scientific Agent Skills free to use?
Yes. It is released under the MIT License, so you can use, modify, and commercialize it with attribution (GitHub, retrieved 2026-10-23).
Which AI agents support these skills?
Any agent that supports the open Agent Skills standard, including Claude Code, Cursor, Codex, Pi, and Antigravity (GitHub, retrieved 2026-10-23).
How is this different from giving my agent access to scientific databases?
Databases give an agent data. Skills give it procedures: the ordered, expert-validated workflows of real labs, plus the context to choose the right one. K-Dense calls skills the "what to do with your tools" layer (K-Dense blog, 2026-06-02).
How many skills does the library include?
177 validated skills as of October 2026 on the live repo, plus access to more than 100 scientific databases. The associated arXiv paper describes a 163-skill snapshot submitted in August 2026 (arXiv:2609.00065, submitted 2026-08-30).