searching-literature_skill

Overview

This skill performs targeted PubMed searches with keyword optimization, result parsing, and metadata extraction to produce a clean initial set of papers for review. It balances broad coverage with focused filters so you find relevant literature without overwhelming volume. Outputs are ready for downstream relevance evaluation and tracking.

How this skill works

It parses the user query to extract keywords, data types (e.g., IC50, methods), constraints (date range, journals, authors), and synonyms. It builds PubMed queries using Boolean operators and field tags, executes searches via the NCBI E-utilities, retrieves PMIDs, and fetches metadata (title, authors, journal, date, abstract, DOI). Results are returned as structured paper objects and saved for follow-up processing.

When to use it

• Starting a new literature review or research question
• User asks “find papers about X” or requests evidence for a claim
• Gathering an initial corpus of papers for manual or automated screening
• Searching for specific methods, compounds, diseases, or metrics (e.g., IC50)
• Preparing inputs for downstream relevance-evaluation or data-extraction skills

Best practices

• Extract synonyms and include OR groups to avoid missing relevant papers
• Start with retmax=100 and iterate—narrow with AND or field tags if too many results
• Respect PubMed rate limits; use ~500 ms delay for single-agent scripts
• If abstracts are missing, use efetch instead of esummary to retrieve full abstracts
• Use DOI where available; fall back to PMID for older papers or missing DOIs

Example use cases

• Find recent and classic papers on BTK inhibitors and selectivity, collecting IC50 reports
• Compile a set of methods papers for a protocol comparison across years
• Search for clinical trials and safety data on a specific compound or drug class
• Create an initial corpus for systematic review screening with metadata and identifiers
• Resolve DOIs for a list of PMIDs to prepare citation lists or DOI-based lookups