arxiv-paper-search

name	arxiv-paper-search
description	Search arXiv papers by keyword, author, category, and date, then return ranked results with links, abstracts, and citation metadata. Use when a user asks for arXiv paper discovery, latest papers, related work curation, author/category monitoring, or Korean prompts such as "arxiv 논문 검색", "최신 arxiv 논문", "arxiv 추천", and "관련 연구 찾아줘".

ArXiv Paper Search

Find and filter arXiv papers with reproducible API queries and explicit date boundaries. Use this skill whenever fresh arXiv metadata is required.

Workflow

1. Resolve the script path for cross-session use.

• Prefer "$CODEX_HOME/skills/arxiv-paper-search/scripts/search_arxiv.py".
• Fallback to skills/arxiv-paper-search/scripts/search_arxiv.py for repository-local use.

2. Capture user intent.

• Extract topic keywords, date window, category/author filters, and result count (N).
• If N is not specified, default to 10 and explicitly report the assumption.
• Convert relative expressions ("today", "this week", "last 30 days") into exact calendar dates in the final response.

3. Build a structured query.

• Prefer --query, --author, and --category.
• Use --raw-query only for exact arXiv syntax.
• Use --days N for rolling windows, or --from-date/--to-date for fixed windows.
• See references/arxiv-query-guide.md for syntax patterns and categories.

4. Execute the script.

• Run with --json when the result will be post-processed.
• Add --include-bibtex when users need references.
• Use --require-pdf, --dedupe-by-title, and --max-abstract-chars when cleaning large result sets.

5. Summarize for the user.

• Report title, authors, published date, primary category, and URL.
• Mention selected filters, date window, and any dropped-result counts.
• Never fabricate papers, metadata, or citations.

6. Handle errors explicitly.

• If API calls fail, report exact stderr and retry guidance.
• If no results are found, suggest one or two query refinements.

Commands

# Resolve a reusable path first (global skill install -> local fallback)
SKILL_ROOT="${CODEX_HOME:-$HOME/.codex}/skills/arxiv-paper-search"
if [ ! -f "$SKILL_ROOT/scripts/search_arxiv.py" ]; then
  SKILL_ROOT="skills/arxiv-paper-search"
fi

# Latest papers by topic (machine-readable output)
N=20
python3 "$SKILL_ROOT/scripts/search_arxiv.py" \
  --query "vision-language model" \
  --sort-by submittedDate --sort-order descending \
  --days 30 \
  --max-results "$N" \
  --dedupe-by-title \
  --json

# Category + absolute date window + BibTeX
python3 "$SKILL_ROOT/scripts/search_arxiv.py" \
  --category cs.LG \
  --query "korean morphology" \
  --from-date 2025-01-01 \
  --to-date 2025-12-31 \
  --max-results 10 \
  --require-pdf \
  --max-abstract-chars 600 \
  --include-bibtex

Table Output (Variable N)

Use the user-requested paper count directly via --max-results N.

N=12
python3 "$SKILL_ROOT/scripts/search_arxiv.py" \
  --query "korean morphological analyzer" \
  --category cs.CL \
  --days 365 \
  --sort-by submittedDate \
  --max-results "$N" \
  --dedupe-by-title \
  --json

When summarizing, provide a Markdown table with one row per returned entry:

| # | Title | Authors | Published | Category | URL |
|---|---|---|---|---|---|
| 1 | ... | ... | 2026-02-17 | cs.CL | https://arxiv.org/abs/... |

Resources

• Search tool: scripts/search_arxiv.py
• Query and category reference: references/arxiv-query-guide.md