Scientific Skills for Claude

Natural language scientific literature search skills with semantic search over PubMed, arXiv, ChEMBL, DrugBank, bioRxiv, medRxiv, clinical trials, and more with zero API complexity.

Why These Skills Are Powerful

• No API Parameter Parsing: Just pass natural language queries - no complex search syntax or parameters needed
• Semantic Search: Understands meaning and context, not just keyword matching
• Full-Text Access: Returns complete article content, not just abstracts
• Image Links: Includes figures and images from papers
• Comprehensive Coverage: Access to ALL scientific literature across multiple databases

Example

Available Skills

Individual Data Sources (9 skills)

| Skill | Data Source | Coverage | |-------|-------------|----------| | pubmed-search | PubMed | Biomedical and life sciences literature | | arxiv-search | arXiv | Physics, mathematics, computer science, quantitative biology preprints | | biorxiv-search | bioRxiv | Biology preprints | | medrxiv-search | medRxiv | Medical and health sciences preprints | | chembl-search | ChEMBL | Bioactive molecules with drug-like properties | | drugbank-search | DrugBank | Comprehensive drug and drug target database | | clinical-trials-search | ClinicalTrials.gov | Clinical trials registry | | drug-labels-search | FDA Drug Labels | Official FDA drug information | | open-targets-search | Open Targets | Drug targets and disease associations | | patents-search | Patent Databases | Global patent filings |

Aggregated Skills (3 skills)

| Skill | Combined Sources | Best For | |-------|------------------|----------| | literature-search | PubMed + arXiv + bioRxiv + medRxiv | General scientific literature review | | biomedical-search | PubMed + bioRxiv + medRxiv + ClinicalTrials + Drug Labels | Medical and clinical research | | drug-discovery-search | ChEMBL + DrugBank + Drug Labels + Open Targets | Drug discovery and development |

Installation

Via skills.sh (Recommended)

npx skills i yorkeccak/scientific-skills

Or browse individual skills at skills.sh - search for "yorkeccak"

Example: skills.sh/yorkeccak/scientific-skills/biomedical-search

Supports installation to multiple agents:

• Antigravity
• Claude Code
• Clawdbot
• Codex
• Cursor
• Droid
• Gemini CLI
• GitHub Copilot
• Goose
• Kilo Code
• Kiro CLI
• OpenCode
• Roo Code
• Trae
• Windsurf

Via Claude Plugin

/plugin install yorkeccak/scientific-skills

Manual Installation

# Clone the repository
cd ~/dev
git clone https://github.com/yorkeccak/scientific-skills.git

# Add as a local plugin
/plugin add ~/dev/scientific-skills

Quick Start

1. Get an API Key for the Valyu search

There is $10 free credits: platform.valyu.ai

2. First Use (Automatic Setup)

Just start using any skill! Claude will automatically:

1. Detect that no API key is configured
2. Ask you to paste your API key
3. Save it to ~/.valyu/config.json
4. Run your search

You: Search PubMed for CRISPR advances in 2024
Claude: "To search PubMed, I need your Valyu API key from https://platform.valyu.ai"
You: val_abc123...
Claude: [saves key and runs search automatically]

3. Manual Setup (Optional)

If you prefer to set up manually:

Environment Variable (recommended):

# For Zsh (macOS default)
echo 'export VALYU_API_KEY="your-api-key-here"' >> ~/.zshrc
source ~/.zshrc

# For Bash
echo 'export VALYU_API_KEY="your-api-key-here"' >> ~/.bashrc
source ~/.bashrc

Config File:

mkdir -p ~/.valyu
echo '{"apiKey": "your-api-key-here"}' > ~/.valyu/config.json

Output Format

All skills return consistent JSON output:

{
  "success": true,
  "type": "pubmed_search",
  "query": "your query",
  "result_count": 10,
  "results": [
    {
      "title": "Article Title",
      "url": "https://...",
      "content": "Full article text with figures...",
      "source": "pubmed",
      "relevance_score": 0.95,
      "images": ["https://example.com/figure1.jpg"]
    }
  ],
  "cost": 0.025
}

Processing Results with jq

# Get article titles
scripts/search "query" 10 | jq -r '.results[].title'

# Get URLs
scripts/search "query" 10 | jq -r '.results[].url'

# Extract content
scripts/search "query" 10 | jq -r '.results[].content'

# Filter by relevance score
scripts/search "query" 20 | jq '.results[] | select(.relevance_score > 0.9)'

Common Use Cases

Literature Review

Search across all scientific literature:

cd ~/dev/scientific-skills/skills/literature-search
scripts/search "epigenetic modifications in cancer" 50

Drug Discovery

Find compounds and targets:

cd ~/dev/scientific-skills/skills/drug-discovery-search
scripts/search "kinase inhibitors for melanoma" 30

Clinical Research

Find trials and patient outcomes:

cd ~/dev/scientific-skills/skills/biomedical-search
scripts/search "checkpoint inhibitor combinations trials" 40

Patent Research

Search for prior art:

cd ~/dev/scientific-skills/skills/patents-search
scripts/search "CRISPR delivery methods" 25

Gene Function Research

Understand gene roles:

cd ~/dev/scientific-skills/skills/pubmed-search
scripts/search "TP53 tumor suppression mechanisms" 20

Requirements

• Node.js 18+ (uses built-in fetch API)
• Valyu API key from platform.valyu.ai
• No npm packages required - zero external dependencies

Technical Details

Architecture

Each skill consists of:

skill-name-search/
├── SKILL.md              # Skill documentation
└── scripts/
    ├── search            # Bash wrapper
    └── search.mjs        # Node.js implementation

FAQ

What's the difference between individual and aggregated skills?

Individual skills search a single data source (e.g., only PubMed). Aggregated skills search multiple related sources simultaneously for more comprehensive results.

How is this different from regular PubMed search?

• Semantic understanding: Understands query meaning, not just keywords
• Full-text access: Returns complete articles, not just abstracts
• Natural language: No need to learn search syntax
• Unified interface: Same approach across all scientific databases

Can I customize the search parameters?

The skills use natural language queries by design to eliminate API complexity. For advanced filtering, you can:

1. Be more specific in your natural language query
2. Adjust the maxResults parameter
3. Filter results using jq after receiving them

Comparison to Traditional Search

| Feature | Traditional Search | These Skills | |---------|-------------------|--------------| | Query Format | Complex syntax, Boolean operators | Natural language | | Search Type | Keyword matching | Semantic understanding | | Content Returned | Abstracts, metadata | Full-text with images | | API Complexity | High - learn each API | Zero - same interface | | Setup Time | Hours per database | Minutes for all | | Multiple Sources | Write separate integrations | Single unified approach |

Contributing

Welcome all PRs!

License

MIT License - See individual skill files for details.