Claude Scientific Skills
by K-Dense-AI
Run a full drug discovery pipeline -- from querying ChEMBL for bioactivity data through molecular docking with DiffDock against AlphaFold structures -- entirely from natural language prompts inside your AI coding assistant. Claude Scientific Skills is a collection of 170+ agent skill files that teach Claude Code, Cursor, Gemini CLI, and Codex how to operate scientific Python packages, query research databases, and execute multi-step analysis workflows across biology, chemistry, medicine, physics, engineering, and finance. The collection covers an unusually broad surface area for a single skill repository. It wraps over 60 Python packages (RDKit, Scanpy, BioPython, PyTorch Lightning, pyOpenMS, and dozens more) with pre-documented implementation patterns that agents can follow without hallucinating API calls. It provides direct access to 250+ databases including PubMed, UniProt, Ensembl, ChEMBL, PubChem, ZINC, ClinicalTrials.gov, ClinVar, and COSMIC. And it includes 15+ integrations with scientific platforms like Benchling, DNAnexus, LatchBio, OMERO, and Protocols.io for lab-connected workflows. Installation is straightforward: clone the repository and copy the skill folders into your agent's skills directory (~/.claude/skills/ for Claude Code, ~/.cursor/skills/ for Cursor). No configuration files, no API keys for the skills themselves, no build step. The agent auto-discovers installed skills and applies them when it detects relevant scientific questions. Each skill folder contains a SKILL.md with structured instructions, working code examples, and best practices that guide the agent through correct package usage. The real value shows up in multi-step scientific workflows. A single conversation can chain together database queries, statistical analysis, molecular simulations, and publication-quality visualization -- tasks that would normally require switching between multiple specialized tools and writing substantial boilerplate code. The repository is maintained by K-Dense AI and released under the MIT license, with a commercial hosted version (K-Dense Web) available for teams that want cloud GPU access and additional exclusive skills beyond the 170 open-source ones.
Installation
Key Features
- ✓170+ skill files that teach AI agents to correctly use scientific Python packages like RDKit, Scanpy, BioPython, and pyOpenMS without hallucinating API calls or inventing nonexistent functions
- ✓Direct access to 250+ research databases including PubMed, UniProt, ChEMBL, PubChem, ClinicalTrials.gov, and COSMIC through pre-documented query patterns the agent follows step by step
- ✓Multi-step scientific workflow execution that chains database queries, statistical analysis, molecular simulations, and visualization in a single conversation -- replacing hours of manual tool-switching
- ✓Cross-platform compatibility with Claude Code, Cursor, Gemini CLI, and Codex through a simple copy-to-skills-directory installation that requires no configuration or API keys
- ✓15+ integrations with scientific platforms like Benchling, DNAnexus, LatchBio, and Protocols.io that connect agent workflows directly to lab information systems and experimental data
- ✓35+ analysis and communication tools covering literature review, scientific writing, statistical analysis, network analysis, and publication-quality figure generation
- ✓Structured SKILL.md files with working code examples and best practices for each domain, giving agents concrete implementation patterns instead of vague instructions
Use Cases
- →A computational biologist needs to analyze single-cell RNA-seq data: load 10X Genomics output, run quality control, integrate with public CellXGene Census datasets, identify cell types, perform differential expression analysis, and infer gene regulatory networks -- all orchestrated through natural language prompts to their AI assistant
- →A medicinal chemist exploring lead optimization queries ChEMBL for bioactivity data on a target, generates molecular analogs with RDKit, filters candidates by ADMET properties, docks top hits against AlphaFold-predicted structures using DiffDock, and ranks results with DeepChem scoring models
- →A clinical researcher interpreting genomic variants parses VCF files with pysam, annotates variants through Ensembl VEP, cross-references ClinVar and COSMIC databases for clinical significance, checks pharmacogenomic interactions via ClinPGx, and generates structured clinical reports
- →A materials scientist investigating crystal structures retrieves data from crystallographic databases, runs computational chemistry calculations, generates phase diagrams, and produces publication-ready visualizations -- without writing boilerplate code for each tool
- →A research team preparing a grant application uses the literature review and scientific writing skills to systematically search PubMed, summarize relevant papers with citations, draft methodology sections, and generate figures that illustrate their proposed experimental approach