arXiv Research Skill

Search, retrieve, and analyze papers from arXiv.org.

Quick Start

bash

# Search recent papers on a topic
python .agents/skills/arxiv/arxiv_cli.py search -q "hypergraph transformer" -n 10 --months 18

# Filter by ML categories
python .agents/skills/arxiv/arxiv_cli.py search -q "LLM reasoning" -c cs.LG -c cs.AI -n 15

# Get paper details (includes HTML link)
python .agents/skills/arxiv/arxiv_cli.py get -i 2211.09590

# Download PDF
python .agents/skills/arxiv/arxiv_cli.py download -i 2211.09590 -o ./papers/

Commands

Search Papers

bash

python .agents/skills/arxiv/arxiv_cli.py search \
  --query "graph neural networks" \
  --max-results 10 \
  --category cs.LG \
  --months 12

Options:

Option	Short	Description
`--query`	`-q`	Search query (required)
`--max-results`	`-n`	Max results (default: 10)
`--sort-by`	`-s`	`relevance`, `date`, `lastUpdatedDate`
`--category`	`-c`	Filter by category (repeatable)
`--months`	`-m`	Papers from last N months
`--since`		Filter after date (YYYY-MM-DD)
`--until`		Filter before date (YYYY-MM-DD)
`--smart`		Use LLM to translate natural language to arXiv query

Examples:

bash

# Recent hypergraph papers in ML
python .agents/skills/arxiv/arxiv_cli.py search -q "hypergraph" -c cs.LG --months 18

# Attention papers from 2024
python .agents/skills/arxiv/arxiv_cli.py search -q "attention mechanism" --since 2024-01-01

# Multiple categories
python .agents/skills/arxiv/arxiv_cli.py search -q "transformer" -c cs.CV -c cs.LG -n 20

# Smart mode - natural language query (uses scillm for translation)
python .agents/skills/arxiv/arxiv_cli.py search --smart \
  -q "papers on hypergraph transformers for machine learning" \
  --months 18 -n 10

Smart Query Mode

The --smart flag uses scillm to translate natural language to arXiv query syntax:

bash

# Input: natural language
python arxiv_cli.py search --smart -q "recent work on LLM reasoning and chain of thought"

# LLM translates to:
# (ti:LLM OR ti:language model) AND (ti:reasoning OR abs:chain of thought)

Requires: CHUTES_API_KEY and CHUTES_MODEL_ID environment variables.

Get Paper Details

bash

python .agents/skills/arxiv/arxiv_cli.py get --paper-id 2301.00001

Returns full metadata including html_url for ar5iv HTML version.

Download PDF

bash

python .agents/skills/arxiv/arxiv_cli.py download \
  --paper-id 2301.00001 \
  --output ./papers/

List Categories

bash

python .agents/skills/arxiv/arxiv_cli.py categories

Common categories:

•cs.LG - Machine Learning
•cs.AI - Artificial Intelligence
•cs.CL - Computation and Language (NLP)
•cs.CV - Computer Vision
•cs.NE - Neural and Evolutionary Computing
•stat.ML - Machine Learning (Stats)

Output Format

All commands output JSON:

json

{
  "meta": {
    "query": "hypergraph transformer",
    "count": 5,
    "took_ms": 234,
    "filters": {"categories": ["cs.LG"], "since": "2024-07-01"}
  },
  "items": [
    {
      "id": "2211.09590v5",
      "title": "Hypergraph Transformer for Skeleton-based Action Recognition",
      "abstract": "...",
      "authors": ["Author 1", "Author 2"],
      "published": "2022-11-17",
      "pdf_url": "https://arxiv.org/pdf/2211.09590.pdf",
      "abs_url": "https://arxiv.org/abs/2211.09590v5",
      "html_url": "https://ar5iv.org/abs/2211.09590",
      "categories": ["cs.CV"],
      "primary_category": "cs.CV"
    }
  ],
  "errors": []
}

Key URLs:

•pdf_url - Direct PDF download
•abs_url - arXiv abstract page
•html_url - ar5iv HTML rendering (better for reading, no PDF needed)

Common Research Queries

bash

# "Find all papers on X from the past Y months"
python .agents/skills/arxiv/arxiv_cli.py search -q "hypergraph transformer" --months 18 -n 20

# "Find ML papers on X"
python .agents/skills/arxiv/arxiv_cli.py search -q "knowledge distillation" -c cs.LG -c cs.AI -n 15

# "What's new in X this year?"
python .agents/skills/arxiv/arxiv_cli.py search -q "mixture of experts" --since 2025-01-01 -s date

Integration with Memory

After finding useful papers:

bash

# Search for papers
python .agents/skills/arxiv/arxiv_cli.py search --query "..." > papers.json

# Ingest into memory (if using memory project)
memory-agent workspace-ingest --source papers.json --scope arxiv

Rate Limits

arXiv API has rate limits (~30 requests/minute). The CLI handles this automatically with backoff.

No API Key Required

arXiv API is free and open - no authentication needed.