arxiv-search

name	arxiv-search
description	Search arXiv with structured results (id, title, authors, date, categories, abstract), grade relevance, and record kept papers in the literature store.
stack	sci

arXiv Search

Turn a research question into a short list of graded, stored papers.

When to Use

• The user asks about a paper, technique, or research area ("is there work on X?", "what's the state of Y?").
• A literature review, related-work section, or verification that a claim matches published work.
• You need authors, dates, or the abstract of a specific arXiv paper.

Procedure

1. Call the tool if installed: the arxiv-search user tool (see packages/coding-agent/scripts/arxiv-search.mjs to install it). Otherwise use the one-liner fallback: curl -s "https://export.arxiv.org/api/query?search_query=all:QUERY&start=0&max_results=8&sortBy=relevance" | grep -o "<title>[^<]*\|<summary>[^<]*" — or run the script directly: node packages/coding-agent/scripts/arxiv-search.mjs --query "QUERY" [--max 8] [--sort submittedDate] [--from YYYY-MM-DD].
2. Read the results: skim titles + abstracts. Keep 1-5 that answer the question; drop the rest.
3. Grade + record: for each kept paper, add it to the literature store (action=add) with the arXiv url, a one-line takeaway in notes, status: to-read (or reviewed if you read the abstract carefully), and a grade: A peer-reviewed/replicated, B single study, C preprint, D unverified. arXiv papers default to C unless you have stronger evidence.
4. Answer the user with the kept papers: title, authors, date, one-line finding, and the link. Cite only what the tool actually returned — never invent a paper.

Pitfalls

• Do not dump all results; the user wants 1-5 relevant ones.
• Relevance sort != recency. For "what's new" use submittedDate.
• An abstract is not the paper. Grading stays C (preprint) until you verify beyond the abstract.
• Do not fabricate arxivIds or URLs — copy them from the tool output.