Home

Create a Claude Skill for Medical Research

Keeping up with medical research means wading through dense scientific papers filled with jargon. A Claude Skill is like having a research assistant that reads hundreds of papers and explains them in plain English - instead of spending hours searching PubMed and decoding technical language, you ask a simple question and get a clear summary based on the latest science.

In this tutorial, you’ll create a Claude Skill that searches peer-reviewed medical literature. We first create a Python script to query PubMed. The best part? You can ask Claude to do all the work! Claude knows how to write Skills, too.

Key Concepts

What You’ll Need

Step 1: Navigate to Project Folder

If you completed the slash commands tutorial, the test_claude folder already exists. These commands work either way.

Windows (PowerShell):

Mac:

The -p (Mac) and -Force (Windows) flags create the folder if it doesn’t exist, or simply do nothing if it already exists.

Step 2: Start Claude Code

Type this command:

claude
Log in with your Claude subscription following the installation tutorial (Windows Mac). After logging in, you’ll see a welcome message and the Claude Code prompt.

Step 3: Let Claude Edit Files Without Asking

Look at the status line just below the Claude Code prompt:

This allows Claude to create and modify multiple files without asking for permission each time. This is essential for building skills which involve creating many files.

Step 4: Create the PubMed Search Script

Copy and paste this prompt:

Write a Python script called pubmed_search.py that:
- Takes a search query as command line argument
- Retrieves up to 10 recent papers
- Returns PMID, title, authors, journal, year, abstract preview, and URL

Claude creates the script. It installs the required Biopython package, which includes an Entrez module for interacting with PubMed. Review the output to see the script structure.

Step 5: Test the Script

Ask Claude to test it:

Test the script with the query "immunotherapy breast cancer"

Claude runs the script and displays 10 recent papers about immunotherapy for breast cancer. You’ll see titles, authors, abstracts, and PubMed links.

Note: The search finds thousands of papers but retrieves only the 10 most recent for quick review.

Step 6: Create the Medical Research Skill

Now we’ll package everything into a reusable skill. Copy and paste:

Create a Claude Skill called "medical-research" that:
- Takes on a medical question
- Designs queries to retrieve PubMed abstracts
- Creates plain-language summaries accessible to non-scientists
- Includes the pubmed_search.py script inside the skill folder

Claude creates the skill structure. The exact files can vary, for example:

This takes 2-3 minutes as Claude writes comprehensive documentation.

Step 7: Test the Skill

First type /reload-skills so Claude Code loads the new Skill. Then ask a research question in plain language:

Can I lose weight on a keto diet?

Claude automatically:

The response includes sections like “How It Works,” “Research Findings,” “Important Considerations,” and “The Bottom Line.”

Step 8: Test Another Question

Try another research topic:

Do vaccines cause autism?

The response explains the scientific consensus, the origin of the myth, and real-world consequences.

Step 9: Explore the Skill Files (Optional)

Open VS Code to see the skill structure:

Windows:

Mac:

In VS Code:

Note: The skill is completely self-contained - all code and documentation live in one folder.

Step 10: Installing Skills from Anthropic’s Repository

Anthropic maintains a repository of pre-built skills that you can install and use instantly.

These document skills help Claude read and write Word, PDF, PowerPoint, and Excel files. Claude copies them into your .claude/skills/ folder.

Other available skills from Anthropic’s repository:

Next Steps

Troubleshooting

Workflow Overview

Skills vs Slash Commands

Feature Skills Slash Commands
What it contains Prompts, code, data, and documentation Prompts only
Activation Automatically when relevant, or manually with /skill-name Manually triggered with /command
Use case Complex workflows requiring code execution Quick prompt templates and workflows
File location .claude/skills/skill-name/ .claude/commands/
Structure Multiple files (SKILL.md, README.md, code files) Single .md file per command

Key Takeaway: Slash commands just inject prompts from the UI. It is useful, but limited.


Created by Steven Ge on December 13, 2025.