| name | arxiv-api-search |
| description | Direct arXiv API search with Python filtering and SQLite storage. Use for real-time query of recent papers without local database. |
arXiv API Search
Search arXiv via API. Results are cached in SQLite for offline re-query.
Usage
Python: python3 (fallback: ask user for path if not found)
# Basic search
python3 <skill_path>/scripts/arxiv_search.py --query "your search terms" --max 20
# With category filter
python3 <skill_path>/scripts/arxiv_search.py --query "surgical navigation" --category "cs.RO" --max 20
# Recent papers only
python3 <skill_path>/scripts/arxiv_search.py --query "deep learning" --recent 3
# Brief mode (id + title + date only, saves tokens for large batch screening)
python3 <skill_path>/scripts/arxiv_search.py --query "..." --brief
# Query local database (no API call)
python3 <skill_path>/scripts/arxiv_search.py --db-query "keyword"
# Markdown output
python3 <skill_path>/scripts/arxiv_search.py --query "..." --format markdown
Shell Compatibility
IMPORTANT: The --query value may contain parentheses and special characters that are interpreted differently across shells. You MUST follow these rules when generating commands:
Parameters
| Param | Description |
|---|
--query, -q | Search query (supports arXiv syntax: all:, ti:, abs:, au:, cat:) |
--max, -m | Max results, default 20 |
--category, -c | Category filter (e.g. cs.RO, cs.CV, cs.AI, cs.LG, eess.IV) |
--recent | Filter to last N years |
--brief | Minimal output: arxiv_id, title, published only |
--format, -f | json (default, compact) or markdown |
--db-query, -d | Search local cache instead of API |
--session-id, -s | Group results by session |
Query Syntax
Simple keywords work best and avoid shell escaping issues:
deep learning surgical navigation # plain keywords (recommended, most compatible)
Advanced syntax (wrap in single quotes on PowerShell):
all:(keyword1 OR keyword2) # search title + abstract
ti:(exact title phrase) # title only
cat:(cs.RO) AND all:(navigation) # category + keyword
au:(Smith) AND ti:(robot) # author + title
Output Fields
Default: arxiv_id, title, abstract, published, categories
Brief mode: arxiv_id, title, published
Authors are stored in database but excluded from output to save tokens.
Notes
- Results sorted by submission date (newest first)
- arXiv rate limit: ~1 request per 3 seconds. Script has built-in retry with backoff.
- Database:
<skill_path>/data/arxiv_cache.db