literature-review

name	literature-review
description	Conduct a rigorous scientific literature review using arXiv and web search. Use when surveying prior work, finding papers, identifying gaps, or summarizing the state of a research area.

Literature Review

A literature review establishes what is known, how well it is known, and what is not known. Follow this workflow when researching a scientific question.

Workflow

1. Scope the question. Write down the research question, the target domain, and inclusion criteria (time range, venue types, key terms). State what kind of evidence would count as an answer.

2. Search systematically. Use the arxiv tool with targeted queries (ti:, abs:, au:, cat:), the websearch tool for broader discovery, and webfetch for specific papers or venues. Try multiple search phrasings. Record each query and its results in research/review/ so the search is reproducible.

3. Collect primary sources. Fetch full abstracts (and full text when available) for the most relevant papers. Save structured notes per paper: citation, claims, methods, evidence strength, limitations, open questions.

4. Synthesize, don't stack. Organize findings by theme, approach, or debate — not by author. Compare competing claims explicitly: what evidence supports each side, and where do they disagree?

5. Identify gaps. End with concrete open questions: what has no direct evidence, what is contested, and what claims would settle the question?

Output

• research/review/ directory with: queries.md (search log), notes/*.md (per-paper notes), and synthesis.md (the final review).
• The synthesis must separate established facts from interpretations and cite every nontrivial claim.

Rules

• Prefer primary sources (arXiv preprints, papers) over secondary summaries.
• Note publication dates; do not treat old results as current.
• Distinguish "no evidence" from "evidence against".
• Never fabricate citations or results. If a claim cannot be traced, say so.