lit-review

name: lit-review description: Conducts a structured literature review with search strategy, synthesis, and gap identification. Use when surveying a research area. argument-hint: "" allowed-tools: Bash, Read, Write, Edit, Glob, Grep, WebSearch, WebFetch version: 1.0.0 workflow_stage: writing tags: - literature-review - synthesis - citations - research-gaps

Structured Literature Review

Conduct a systematic literature review: define scope, design search strategy, build a paper inventory, synthesize findings, and identify research gaps.

Arguments

• $ARGUMENTS — a research question or topic to survey (e.g., "convergence in regional GDP per capita", "effects of trade liberalization on inequality", "spatial spillovers in economic growth")

Steps

1. Parse the research question or topic from the arguments.

2. Define scope. Ask the user to clarify:

   • Subfield and disciplinary boundaries (e.g., growth economics, trade, labor)
   • Time frame for included papers (e.g., 2000–present, or seminal papers from any era)
   • Key journals to prioritize (e.g., AER, QJE, ReStud, JDE, WBER)
   • Seed papers — 2–3 known papers to anchor the search
   • Inclusion/exclusion criteria (e.g., empirical only, specific methods, specific geographies)

3. Design search strategy. Generate structured search strings for:

   • Google Scholar: combine key terms with Boolean operators ("economic convergence" AND "panel data" AND (regional OR subnational))
   • NBER Working Papers: topic-based search for recent unpublished work
   • EconLit / IDEAS/RePEc: JEL code filters (e.g., O47 for economic growth, F14 for trade)
   • Document all search queries used — the strategy must be replicable

4. Build paper inventory. For each paper found, record a row in a summary table:

   #	Authors (Year)	Journal	Method	Key Finding	Relevance (1–5)

   Prioritize papers with relevance score >= 3. Aim for 15–30 papers for a focused review, 50+ for a comprehensive survey.

5. Synthesize. Group papers by theme and identify:

   • Consensus findings: What does the weight of evidence support?
   • Contested findings: Where do papers disagree, and why? (different samples, methods, time periods)
   • Methodological evolution: How have methods improved over time? (e.g., from OLS to DiD to staggered DiD)
   • Geographic/temporal gaps: Which regions, countries, or time periods are understudied?

6. Generate output. Save to references/lit-review-<topic-slug>.md with this structure:

   # Literature Review: <Topic>
   
   **Date:** YYYY-MM-DD
   **Scope:** <subfield, time frame, criteria>
   
   ## Search Strategy
   
   - Google Scholar: `<query>`
   - NBER: `<query>`
   - EconLit JEL codes: `<codes>`
   - Seed papers: <list>
   
   ## Paper Inventory
   
   | # | Authors (Year) | Journal | Method | Key Finding | Relevance |
   |---|----------------|---------|--------|-------------|-----------|
   
   ## Synthesis by Theme
   
   ### Theme 1: <name>
   <discussion of findings, agreements, disagreements>
   
   ### Theme 2: <name>
   ...
   
   ## Identified Gaps
   
   - <gap 1>
   - <gap 2>
   
   ## Implications for This Project
   
   <how the review informs the current research — which methods to use, which gaps to fill, how to position the contribution>
   

7. Integration with other skills. For each high-relevance paper (score >= 4):

   • Offer to run /project:cite to add it to references.bib
   • Offer to run /project:literature-note to create a detailed annotation

Error handling

• If the topic is too broad (e.g., "economics"), ask the user to narrow the scope.
• If no seed papers are provided, suggest 2–3 seminal papers based on the topic and ask for confirmation.

Common Pitfalls

• Confirmation bias: Actively search for papers that contradict the expected finding — a review that only cites supporting evidence is incomplete
• Ignoring working papers: Recent NBER/CEPR/IZA working papers often contain the most current methods and findings; don't rely solely on published articles
• Not documenting the search strategy: Without documented queries, the review is not replicable. Record every search string and database used.
• Selective citation: Don't cherry-pick one estimate from a paper that reports many — note the range of results
• Missing methodological context: When comparing findings across papers, note differences in identification strategy, sample, and time period that explain divergent results