---
name: literature-review
description: Synthesize a bounded body of literature into a source-traceable narrative organized by questions, mechanisms, methods, findings, and unresolved disagreements.
---

# Literature review

Use this Academic Research Assistant skill for narrative, thematic, conceptual, or methodological literature reviews. Follow the [research contract](../../docs/RESEARCH-CONTRACT.md). A review should explain relationships among works and the implications of their differences; an annotated list is useful groundwork but is not yet a synthesis.

## Inputs and coverage

Obtain the review question, audience, discipline, length or format, available corpus, search record, and intended purpose. Establish whether the review maps a field, explains a dispute, compares methods, develops a conceptual framework, or supports a specific project. State the temporal, linguistic, disciplinary, and publication boundaries.

If the corpus is user-supplied, distinguish a review of that corpus from a review of the entire field. Use [literature search](../literature-search/SKILL.md) when discovery is part of the task. If the user needs exhaustive, protocol-driven eligibility and reproducible screening, route that component to [systematic review](../systematic-review/SKILL.md). Do not relabel a convenience sample as a systematic search.

## Build the synthesis

1. Define the review's organizing questions before writing sections. Useful organizing dimensions include theory, mechanism, design, population, context, measurement, parameter regime, and historical development. Choose the structure that exposes the substantive issue instead of reproducing the order of retrieved papers.
2. Create a source matrix with the question, approach, context, principal findings, limitations, and relevant source locations for each work. Distinguish the authors' own results from claims inherited through citations. Use [evidence extraction](../evidence-extraction/SKILL.md) for a larger or more structured corpus.
3. Group comparable claims and investigate disagreement. Differences can reflect definitions, selection, measurement, analytic assumptions, contexts, or uncertainty rather than direct contradiction. Identify which explanation is supported and which remains a possibility.
4. Weigh evidence using design and inference rather than counting positive papers. Multiple reports using the same dataset are not independent confirmations. A highly cited work may be influential without being the strongest evidence for a specific claim.
5. Write synthesis paragraphs around an assertion, its evidence, relevant variation or counterevidence, and the implication for the review question. Use individual-paper summaries when a distinctive method or result needs explanation, not as the default paragraph pattern.
6. Identify gaps precisely. Distinguish an unstudied setting, uncertain measurement, an unresolved theoretical incompatibility, a weak identification strategy, and a reproducibility failure. Explain why filling the gap would change understanding or decisions. Absence from one search is insufficient to claim that no research exists.
7. Conclude with what is established, what depends on conditions, and what remains unresolved. A proposed research direction should follow from the synthesis and its limits rather than a promise of originality.

## Research-mode distinctions

Quantitative findings require attention to effect definitions, populations, uncertainty, and design. Do not compare estimates with incompatible denominators or interpret significance differences as evidence that effects differ. Qualitative synthesis should retain context, interpretive stance, and the meaning of concepts; do not equate recurrent themes with population frequencies.

For theoretical literature, compare assumptions, definitions, domains, and explanatory or deductive aims before comparing conclusions. For computational research, distinguish algorithmic advances, benchmark choices, implementation quality, datasets, and resource budgets. Apparent improvement may arise from changed evaluation conditions. Mixed bodies of evidence can be integrated by explaining what each contributes without pretending they estimate the same quantity.

## Deliverable and checks

Deliver the review, source matrix, search-coverage note, and a short unresolved-questions list. Adapt headings and depth to the requested document. Preserve editable citations and pass bibliography mechanics to [citations and bibliographies](../citations-and-bibliographies/SKILL.md).

Verify that every substantive source attribution is supported by a consulted location. Recheck quotations and numerical comparisons. Ensure the review includes relevant contrary evidence, represents versions accurately, and labels inference as inference. Mark inaccessible sources and unverified claims explicitly. The final review should make clear whether its conclusions describe the sampled corpus or justify broader statements about the field.