Who is scientific-agent-skills for?

Researchers, students and developers who already use an agent such as Claude Code or Cursor and want it to handle scientific work more reliably; some domain knowledge helps you check the results.

What can you build with scientific-agent-skills?

Ask your agent to screen molecules and run ADMET analysis using RDKit-based skills. Analyze single-cell RNA-seq data with an agent that follows Scanpy conventions. Look up compounds, targets or clinical trials across PubChem, ChEMBL and ClinicalTrials.gov. Draft a literature review or point-by-point reply to reviewer comments with source-traceable writing. Build publication-quality figures, slides or posters from research data.

K-Dense-AI /

scientific-agent-skills

A library of 177 science skills that teach AI agents to use research tools and databases

53.9k stars
6.6k stars since 3 Oct
Python
Updated 4 days ago
Some experience
Agent skills & plugins

Open on GitHub (opens in a new tab) Website (opens in a new tab)

AI agents can call almost any Python package, but they often miss the conventions of a scientific field: which endpoint to query, how to page through results, what checks to run before trusting an output. This repo supplies that know-how as skills. A skill is a folder with a SKILL.md file that an agent reads when a task matches.

The skills fall into a few groups. More than 70 cover Python packages such as RDKit, Scanpy, scikit-learn, pymatgen and Qiskit. A Database Lookup skill covers 80 databases, including PubChem, ChEMBL, UniProt and ClinicalTrials.gov. A few integrate with lab platforms like Benchling and Opentrons. Others handle literature review, scientific writing, peer review, and slides or posters. Some skills also ship scripts, templates and tests.

It follows the open Agent Skills standard, so it works with Cursor, Claude Code, Codex, Google Antigravity and other compatible hosts. It is also packaged as an Agent Plugins bundle. The README says the agent can still use any package on its own; these skills add curated, version-aware guidance for the ones listed. Clinical and animal-welfare skills are described as research aids, not decision tools.

What it does

Broad scientific coverage
177 skills across bioinformatics, cheminformatics, proteomics, clinical research, neuroscience, materials science, physics, geospatial science and more.
Database lookup
One skill documents 80 databases with endpoint selection, pagination and provenance. Separate skills cover DepMap, PrimeKG, AlphaGenome and others.
Package-specific guidance
More than 70 skills give version-aware workflows for packages like RDKit, Scanpy, PyTorch Lightning and GeoPandas.
Lab and writing tools
Integrations for Benchling, Opentrons, Protocols.io and LabArchives, plus skills for literature review, peer review, grant writing and slides.
Tested and portable
Skills with scripts have test suites, the repo runs a security scan, and it works with any host that supports the Agent Skills standard.

What you could build

Ask your agent to screen molecules and run ADMET analysis using RDKit-based skills.
Analyze single-cell RNA-seq data with an agent that follows Scanpy conventions.
Look up compounds, targets or clinical trials across PubChem, ChEMBL and ClinicalTrials.gov.
Draft a literature review or point-by-point reply to reviewer comments with source-traceable writing.
Build publication-quality figures, slides or posters from research data.

Get started

The README excerpt doesn't include install commands. Check the repo's Getting Started section for your agent, then copy the skills (or the plugin) into your host's skills folder and ask the agent to use them on a science task.

Commands are copied from the project's README. Read it on GitHub for the full setup.