| 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
-
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.
-
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.
-
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.
-
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?
-
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.
