literature-review

name	literature-review
description	Search papers via Semantic Scholar and arXiv APIs, synthesize findings across a topic, identify research gaps, and generate BibTeX entries. Use when asked to survey literature, find related work, or explore what exists on a research topic.

Literature Review

Available APIs

Semantic Scholar API (no auth required)

# Search papers
curl -s "https://api.semanticscholar.org/graph/v1/paper/search?query=YOUR_QUERY&limit=10&fields=title,authors,year,abstract,citationCount,venue,url"

# Get paper details by ID
curl -s "https://api.semanticscholar.org/graph/v1/paper/PAPER_ID?fields=title,authors,year,abstract,citationCount,references,citations"

# Search by author
curl -s "https://api.semanticscholar.org/graph/v1/author/search?query=AUTHOR_NAME"

arXiv API

# Search papers (returns Atom XML)
curl -s "http://export.arxiv.org/api/query?search_query=all:YOUR_QUERY&start=0&max_results=10"

# Search by category
curl -s "http://export.arxiv.org/api/query?search_query=cat:cs.RO+AND+all:manipulation&max_results=20"

Relevant categories: cs.RO (robotics), cs.LG (machine learning), cs.CV (computer vision), cs.AI (AI)

Zotero Library (via zotero tool)

The user's Zotero library is accessible through the zotero custom tool (requires Zotero + Better BibTeX running). Use this to:

• Check if papers are already in the library before recommending them
• Retrieve existing cite keys for papers the user already has
• Get BibTeX entries in the user's preferred format (Better BibTeX)
• Access PDF annotations and notes on papers

Actions:

• zotero(action="search", query="...") — search the library
• zotero(action="cite", citekeys=["key1", "key2"]) — get BibTeX entries
• zotero(action="details", citekeys=["key1"]) — get full metadata, notes, and annotations
• zotero(action="collections", citekeys=["key1"]) — get collection membership

Falls back to searching the local BibTeX file (~/Documents/roam/library.bib) when Zotero is offline.

Workflow

1. Scope Definition

Clarify before searching:

• What specific topic or problem?
• Time range? (default: last 3-5 years for methods, all time for seminal works)
• Target venues?
• Known seed papers to start from?

2. Library Check

Before external search, check what the user already has:

• Use zotero(action="search", query="...") with the topic keywords
• Note which relevant papers are already in the library (with their cite keys)
• This avoids recommending papers the user already knows and provides ready-to-use cite keys

3. Search Execution

Run searches via bash using the APIs above:

• Use multiple query variations to catch different terminology
• Filter by citation count for influential papers
• Check both Semantic Scholar (broader coverage) and arXiv (recent preprints)
• Cross-reference results with Zotero to flag papers already in the library

4. Paper Synthesis

For each relevant paper, extract into a table:

Paper	Year	Venue	Key Contribution	Method	Results	Limitations	In Library?

5. Thematic Grouping

Organize papers by approach or theme, not chronologically. Each group gets a paragraph explaining the line of work, common techniques, strengths, and where it falls short.

6. Gap Analysis

Based on the literature:

• What problems remain unsolved?
• Where do methods fail?
• What assumptions are commonly made but rarely questioned?
• What combinations haven't been tried?

7. BibTeX Generation

For papers already in the user's Zotero library, use zotero(action="cite", citekeys=[...]) to get BibTeX in their preferred Better BibTeX format.

For new papers not in the library, provide BibTeX entries:

@inproceedings{author2024title,
  title={Paper Title},
  author={Author, First and Author, Second},
  booktitle={Conference Name},
  year={2024}
}

Output Format

Literature Review: [Topic]

Search Queries Used: [list queries executed] Papers Found: X total, Y highly relevant

Summary Table

Thematic Analysis

Research Gaps

Key Takeaways (3-5 insights)

References (BibTeX)

Important

• Never fabricate paper titles, authors, or venues. If a paper seems relevant but cannot be verified via API, say so explicitly.
• Distinguish between arXiv preprints (not peer-reviewed) and published work.
• Be honest about search limitations. API results may not be exhaustive.