| name | literature-search |
| description | Execute multi-database academic literature search |
| allowed-tools | ArxivSearch SemanticScholarSearch SSRNSearch PaperDownload Read Write Bash |
Literature Search Skill
When the user requests an academic literature search, use this skill.
Steps
-
Parse query and generate search keywords
- Analyze the user's research topic and extract core keywords and related concepts
- Generate multiple sets of search queries (in English) covering:
- Direct search on the core topic
- Related methodology search
- Cross-domain intersection search
- Tailor queries per database (arXiv category codes, Semantic Scholar field filters, etc.)
-
Parallel multi-database search
- Search arXiv (focus on relevant categories such as q-fin, stat.ML, cs.LG)
- Search Semantic Scholar (leverage citation graphs to discover key papers)
- Search SSRN (financial economics and related domains)
- For highly cited papers, traverse citation networks (forward + backward citations)
-
Merge and deduplicate results
- Deduplicate based on DOI or title similarity
- Consolidate metadata from multiple sources into a unified record
-
Rank by citation count and relevance
- Sort results by a weighted combination of citation count and relevance score
- Filter to top-50 core papers
-
Return structured result list
- Each result includes: title, authors, year, venue, abstract, citation count, source database, PDF URL
- Flag open-access availability
- Indicate papers that may require manual download (e.g., gated SSRN papers)
Notes
- Prioritize papers from the last 3 years, but do not ignore foundational work
- Tag each paper with its source database
- If a search direction yields insufficient results, explicitly note "insufficient literature" rather than fabricating entries
- Respect API rate limits: arXiv (~1 request per 3 seconds), Semantic Scholar (configure S2_API_KEY for higher limits)
