Scientific Agent Skills

Overview

Scientific Agent Skills is an open library of scientific procedures for AI agents, created by K-Dense. The current repository lists 177 skills; the associated paper describes 163 skills across 16 areas of practice. Each skill is a directory built around a versioned, human-readable SKILL.md instruction file, often with reference material and runnable scripts. The collection follows the open Agent Skills standard and is also packaged as an Agent Plugins package.

Key Features

• 177 scientific and research skills spanning biology, chemistry, medicine, physics, engineering, Earth science, data analysis, and scientific communication.
• Each skill has a SKILL.md with purpose, workflow, and version metadata, plus optional code examples, reference documentation, executable helpers, or templates.
• Documents 100+ scientific and financial databases, including a Database Lookup skill covering 80 sources such as PubChem, ChEMBL, UniProt, COSMIC, ClinicalTrials.gov, FRED, and USPTO.
• Provides 70+ optimized Python package skills with version-aware workflows for packages such as RDKit, Scanpy, scikit-learn, pydicom, QuTiP, GeoPandas, pymatgen, and Qiskit.
• Follows the Agent Skills standard and works with hosts such as Cursor, Claude Code, Codex, and Google Antigravity; also available as an Agent Plugins package.
• Includes skills for scientific integration platforms such as Benchling, DNAnexus, LatchBio, OMERO, Protocols.io, Open Notebook, Ginkgo Cloud Lab, LabArchives, and Opentrons.

Use Cases

• Guiding AI agents through field-accepted analysis procedures, validation checks, and provenance requirements in scientific workflows.
• Working with specialized scientific libraries, databases, and tools across genomics, cheminformatics, medical imaging, and other research domains.
• Supporting research tasks such as literature review, evidence-traceable scientific writing, grant writing, and confidential peer review assistance.
• Packaging reusable scientific skills for compatible agent hosts through the Agent Skills and Agent Plugins standards.

Getting Started

• Install with the standards-based installer by running npx skills add K-Dense-AI/scientific-agent-skills for supported hosts.
• With GitHub CLI v2.90.0 or later, run gh skill install K-Dense-AI/scientific-agent-skills; optionally target a host with --agent cursor, --agent claude-code, --agent codex, or --agent gemini.
• Browse the illustrated skill guides, skill categories, or complete catalog to choose skills relevant to your work.

Deployment & Requirements

• The npx installer is described for supported Agent Skills hosts including current versions of Claude Code, Claude Cowork, Codex, Gemini CLI, Google Antigravity, and Cursor.
• GitHub CLI installation requires GitHub CLI v2.90.0 or later.

Before You Adopt

• License: MIT. Review its terms before using, modifying, or distributing the project.
• The associated paper reports no task-level evaluation and no host selection rate, so it does not provide task-level performance evidence.
• The paper warns that 29 of 46 documented workflows would exceed the token window if every reference file were loaded.
• Each skill records compatibility requirements and validation scope, and the source advises checking those details before reusing an older workflow because test coverage differs.