Claude Scientific Skills

Claude Scientific Skills

Overview

A comprehensive collection of 139 ready-to-use scientific skills that transform Claude into an AI research assistant capable of executing complex multi-step scientific workflows across biology, chemistry, medicine, and related fields.

When to Use

Invoke this skill when:

• Working on scientific research tasks
• Need access to specialized databases (PubMed, ChEMBL, UniProt, etc.)
• Performing bioinformatics or cheminformatics analysis
• Creating literature reviews or scientific documents
• Analyzing single-cell RNA-seq, proteomics, or multi-omics data
• Drug discovery and molecular analysis workflows
• Statistical analysis and machine learning on scientific data

Quick Start

// Invoke the main skill catalog
Skill({ skill: 'scientific-skills' });

// Or invoke specific sub-skills directly
Skill({ skill: 'scientific-skills/​rdkit' }); // Cheminformatics
Skill({ skill: 'scientific-skills/​scanpy' }); // Single-cell analysis
Skill({ skill: 'scientific-skills/​biopython' }); // Bioinformatics
Skill({ skill: 'scientific-skills/​literature-review' }); // Literature review

Skill Categories

Scientific Databases (28+)

Python Analysis Libraries (55+)

Bioinformatics & Genomics

Cheminformatics & Drug Discovery

Scientific Communication

Clinical & Medical

Laboratory & Integration

Core Workflows

Literature Review Workflow

# 7-phase systematic literature review
# 1. Planning with PICO framework
# 2. Multi-database search execution
# 3. Screening with PRISMA flow
# 4. Data extraction and quality assessment
# 5. Thematic synthesis
# 6. Citation verification
# 7. PDF generation

Drug Discovery Workflow

# Using RDKit + ChEMBL + datamol
from rdkit import Chem
from rdkit.Chem import Descriptors, AllChem

# 1. Query ChEMBL for bioactivity data
# 2. Calculate molecular properties
# 3. Filter by drug-likeness (Lipinski)
# 4. Similarity screening
# 5. Substructure analysis

Single-Cell Analysis Workflow

# Using scanpy + anndata
import scanpy as sc

# 1. Load and QC data
# 2. Normalization and feature selection
# 3. Dimensionality reduction (PCA, UMAP)
# 4. Clustering (Leiden algorithm)
# 5. Marker gene identification
# 6. Cell type annotation

Hypothesis Generation Workflow

# 8-step systematic process
# 1. Understand phenomenon
# 2. Literature search
# 3. Synthesize evidence
# 4. Generate competing hypotheses
# 5. Evaluate quality
# 6. Design experiments
# 7. Formulate predictions
# 8. Generate report

Sub-Skill Structure

Each sub-skill follows a consistent structure:

scientific-skills/
├── SKILL.md                    # This file (catalog/​index)
├── skills/                     # Individual skill directories
│   ├── rdkit/
│   │   ├── SKILL.md           # Skill documentation
│   │   ├── references/        # API references, patterns
│   │   └── scripts/           # Example scripts
│   ├── scanpy/
│   ├── biopython/
│   └── ... (139 total)

Invoking Sub-Skills

Direct Invocation

// Invoke specific skill
Skill({ skill: 'scientific-skills/​rdkit' });
Skill({ skill: 'scientific-skills/​scanpy' });

Chained Workflows

// Multi-skill workflow
Skill({ skill: 'scientific-skills/​literature-review' });
Skill({ skill: 'scientific-skills/​hypothesis-generation' });
Skill({ skill: 'scientific-skills/​scientific-schematics' });

Prerequisites

• Python 3.9+ (3.12+ recommended)
• uv package manager (recommended)
• Platform: macOS, Linux, or Windows with WSL2

Best Practices

1. Start with the right skill: Use the category tables above to find appropriate skills
2. Chain skills for complex workflows: Literature review → Hypothesis → Experiment design
3. Use database skills for data access: Query databases before analysis
4. Visualize results: Use matplotlib/​seaborn/​plotly skills for publication-quality figures
5. Document findings: Use scientific-writing skill for formal documentation

Integration with Agent Framework

Recommended Agent Pairings

Example Agent Spawn

Task({
  task_id: 'task-1',
  subagent_type: 'python-pro',
  description: 'Analyze molecular dataset with RDKit',
  prompt: `You are the PYTHON-PRO agent with scientific research expertise.

## Task
Analyze the molecular dataset for drug-likeness properties.

## Skills to Invoke
1. Skill({ skill: "scientific-skills/​rdkit" })
2. Skill({ skill: "scientific-skills/​datamol" })

## Workflow
1. Load molecular data
2. Calculate descriptors
3. Apply Lipinski filters
4. Generate visualization
5. Report findings
`,
});

Resources

Bundled Documentation

• skills/​*/​SKILL.md - Individual skill documentation
• skills/​*/​references/ - API references and patterns
• skills/​*/​scripts/ - Example scripts and templates

External Resources

• K-Dense AI GitHub
• RDKit Documentation
• Scanpy Documentation
• BioPython Tutorial

Iron Laws

1. ALWAYS query scientific databases (PubMed, ChEMBL, UniProt) before performing any analysis — raw analysis without literature and database context produces uninformed conclusions that duplicate prior work.
2. NEVER perform analysis without documenting all steps (data sources, parameters, library versions, transformations) — undocumented research is irreproducible and cannot be peer-reviewed or extended.
3. ALWAYS chain multiple domain-specific skills for complex workflows — single-tool analysis misses interdependencies across biology, chemistry, and clinical domains.
4. NEVER report findings without statistical validation — scientific claims require appropriately sized samples, validated methods, and quantified uncertainty.
5. ALWAYS visualize intermediate results after each major processing step — data errors and outliers surface in visualizations before propagating silently to final conclusions.

Anti-Patterns

Memory Protocol (MANDATORY)

Before starting: Read .claude/​context/​memory/​learnings.md

After completing:

• New pattern → .claude/​context/​memory/​learnings.md
• Issue found → .claude/​context/​memory/​issues.md
• Decision made → .claude/​context/​memory/​decisions.md

ASSUME INTERRUPTION: If it's not in memory, it didn't happen.

Version History

• v2.17.0 - Current version with 139 skills
• Integrated from K-Dense-AI/​claude-scientific-skills repository

License

MIT License - Open source and freely available for research and commercial use.