Structured Literature Review

Problem

When a topic spans journals, conferences, preprints, and multiple databases, manual reviews often suffer from incomplete search coverage, drifting inclusion criteria, and write-ups that read like paper-by-paper summaries. This skill turns the review into an engineering workflow: the agent first fixes the research question, then applies reproducible rules for screening, extraction, and synthesis.

Workflow and limits

The core path typically includes:
- Use PICO or an equivalent frame to define Population, Intervention, Comparison, and Outcome, plus time, language, and source-type constraints;
- Generate boolean queries across Semantic Scholar, arXiv, PubMed, and ACM DL, recording queries and dates;
- Apply predefined criteria to remove duplicates, retractions, off-topic items, and non-scholarly content, preserving stage counts for a PRISMA-style flow;
- Extract model, method, metrics, limitations, and relevance to the review question into a consistent table;
- Synthesize consensus, disagreement, and gaps by theme rather than listing papers.

It fits academic research, technology landscape analysis, and evidence-based decisions, especially writing tasks that require cited evidence, method comparisons, and explicit evidence gaps. Note: if very few relevant papers exist, reframe as a scoping review or gap analysis; mark preprints as unreviewed; and state non-English source limits in methods.