Academic Literature Review Workbench

Problem Framing

A difficult part of literature review work is not just finding papers, but making the process reproducible: search strings are missing, database coverage is uneven, results are summarized paper by paper instead of by theme, and DOI or citation formatting keeps breaking at the end. literature-review treats this as a structured academic workflow rather than a one-off writing task.

Workflow and Capabilities

The skill organizes review production around a sequence of auditable steps:

• Planning and scoping: define the research question, review type, time range, languages, publication types, and study designs, using PICO where appropriate.
• Multi-database search: combine gget, bioservices, and datacommons-client to query sources such as PubMed, bioRxiv, ChEMBL, KEGG, UniProt, arXiv, and Semantic Scholar, while recording search strings and parameters.
• Screening and synthesis: remove duplicates, screen by title, abstract, and full text, log exclusion reasons, and produce a PRISMA-style flow; then extract design, sample, findings, and limitations, organizing evidence around 3-5 themes.
• Citation validation and output: use verify_citations.py to check DOIs, authors, titles, and publication metadata, format references in styles such as APA, Nature, or Vancouver, and export a structured PDF with generate_pdf.py.

It is most useful as an execution framework for complex reviews where search provenance, citation correctness, and document structure matter. Preprints still require careful validation, PDF generation depends on local pandoc and xelatex environments, and database quality, API availability, and domain fit still require researcher judgment.