Run a literature review from a research question using Claude Code skills
Idea + starter prompt, not run by us yet
For: Researchers and students who need a structured review of papers
A doctor who does research made several Claude Code skills, with Python code behind them that calls free literature APIs. The agent first interviews you to fix the research question, then queries PubMed, OpenAlex and Semantic Scholar, removes duplicates, screens abstracts twice, follows citations, downloads open access PDFs and writes the synthesis.
Steps
- Clone the author's lit-review-agent repo and read its disclaimer. It says to verify all outputs.
- Get a free Semantic Scholar key and add it in the tool's settings yourself. The shared pool often hits rate limits.
- Answer the interview to fix your question, then approve each of the six checkpoints before it moves on.
- Download paywalled PDFs yourself. That manual step assumes macOS.
- Plan for about 4 to 6 sessions, and compare your first results with the repo's worked demo on heart failure care transitions.
Starter prompt
I want a literature review on [research question]. Start by interviewing me to set the question in PICO form: population, intervention, comparison and outcome. Before you search anything, show me your plan, the databases and the search terms. Then wait for my approval at each checkpoint. Work in a new folder called [review folder] and don't change my other files. List any paywalled papers so I can download them myself, and never ask for my library or publisher passwords. For every claim in the synthesis, cite the paper so I can verify it. Say what you could not find or check.
Original post: GitHub · found via agentusecases.ai. Written up in our own words. We haven't run it, so test it on made-up data first.