literature-search

literature-search finds relevant academic papers by querying several scholarly databases and returning a single, structured result set. Semantic Scholar is the primary source (strong for machine-learning and AI work and able to emit BibTeX), arXiv surfaces the latest preprints, OpenAlex adds the broadest free coverage with no API key, and CrossRef provides DOI-based lookup across the widest range of document types. Results are emitted as JSONL carrying title, authors, year, venue, abstract, citation counts, and BibTeX, and are finally presented as a table plus detailed entries keyed for citation.

Use it when finding papers, checking related work, or assembling a bibliography. The workflow expands the user's request into two to four complementary queries, runs Semantic Scholar first, adds arXiv for very recent preprints and optionally OpenAlex for breadth, then merges and deduplicates the JSONL files into one set. Candidates are ranked by a weighted blend of citations (0.3), recency (0.3), venue quality (0.2), and relevance (0.2), with venues bucketed into tiers: Tier 1 includes NeurIPS, ICML, ICLR, ACL, and CVPR, while workshops and arXiv preprints are flagged as preprints.

Helper scripts cover the full pipeline: per-database search, merge and dedup via paper_db.py, BibTeX generation with bibtex_manager.py, and download_arxiv_source.py to pull a paper's source .tex files by title or arXiv id. It sits upstream of related skills for citation management, literature review, and related-work writing, and complements deep-research and novelty-assessment.