| name | lit-search |
| description | Search academic literature using Semantic Scholar API or WebSearch. Returns papers with titles, abstracts, citation counts, and optionally bibtex entries. |
Literature Search
You are performing an academic literature search for the AI Scientist pipeline.
Arguments
The user's message contains the search query. Extract it and search for relevant papers.
Procedure
-
Parse the query from the user's message or arguments.
-
Try Semantic Scholar first by running:
python3 tools/search.py "<query>" --limit 10 --json
This uses the S2 API (with or without S2_API_KEY).
-
If S2 fails or returns no results, fall back to the WebSearch tool:
- Search for:
"<query>" site:arxiv.org OR site:semanticscholar.org
- For each relevant result, extract title, authors, year, and abstract.
-
Present results in a clear format:
- Numbered list with title, authors, year, venue
- Citation count (if available)
- Brief abstract excerpt
- BibTeX entry (if available and requested)
-
If the user asks for bibtex, extract it from the S2 results or generate proper bibtex entries from the search results.
Notes
- Prioritize papers by citation count (most cited first).
- For novelty checking, highlight papers that are closely related to the user's topic.
- If searching for a specific paper by title, use exact title in quotes.
- The
tools/search.py script is located relative to the project root.