arXiv Research — Setup Guide

Source: nousresearch/hermes-agent (77 peak installs) Category: Research / Academic License: MIT · Platforms: Linux, macOS, Windows Dependencies: None (curl only)

Search and retrieve academic papers from arXiv via their free REST API. No API key, no Python packages, no authentication — just curl and the arXiv API. Returns Atom XML with paper metadata, abstracts, and links to full-text PDFs.

---

What It Does

Capability	How
Keyword search	Search by title, abstract, author, or full text
Author lookup	Find all papers by a specific researcher
Category browse	Filter by arXiv category (cs.AI, stat.ML, etc.)
Paper by ID	Retrieve exact paper by arXiv ID
Abstract extraction	Get clean abstract text via web extraction
Full-text access	Fetch PDF via arXiv's CDN

---

Installation

Via skills.sh (Recommended)

npx skills add nousresearch/hermes-agent --skill arxiv

Direct from Hermes Agent Repo

git clone --depth 1 https://github.com/nousresearch/hermes-agent.git /tmp/hermes-agent
cp -r /tmp/hermes-agent/skills/research/arxiv ~/.hermes/skills/

Manual (Single Install)

mkdir -p ~/.hermes/skills/arxiv
curl -o ~/.hermes/skills/arxiv/SKILL.md \
  https://raw.githubusercontent.com/nousresearch/hermes-agent/main/skills/research/arxiv/SKILL.md

---

Usage with Hermes Agent

Quick Reference

Action	Command / Prompt
Search papers	"Search arXiv for GRPO reinforcement learning papers"
Get specific paper	"Get arXiv paper 2402.03300"
Read abstract	"Show me the abstract of the latest RAG survey on arXiv"
Author search	"Find papers by Yann LeCun on arXiv"
Category filter	"Show me recent cs.CL papers about transformer architectures"

API Commands (Behind the Scenes)

# Search papers
curl -s "https://export.arxiv.org/api/query?search_query=all:GRPO+reinforcement+learning&max_results=5&sortBy=submittedDate&sortOrder=descending"

# Get specific paper
curl -s "https://export.arxiv.org/api/query?id_list=2402.03300"

# Search by author
curl -s "https://export.arxiv.org/api/query?search_query=au:lecun&max_results=5"

# Filter by category
curl -s "https://export.arxiv.org/api/query?search_query=cat:cs.AI+AND+all:attention+mechanism&max_results=5"

Clean Output (Parse XML)

The API returns Atom XML. Hermes auto-parses this, but you can also pipe through Python:

curl -s "https://export.arxiv.org/api/query?search_query=all:GRPO&max_results=3&sortBy=submittedDate&sortOrder=descending" | python3 -c "
import sys, xml.etree.ElementTree as ET
ns = {'atom': 'http://www.w3.org/2005/Atom'}
root = ET.fromstring(sys.stdin.read())
for entry in root.findall('atom:entry', ns):
    title = entry.find('atom:title', ns).text.strip().replace('\n', ' ')
    authors = [a.find('atom:name', ns).text for a in entry.findall('atom:author', ns)]
    link = entry.find('atom:id', ns).text
    print(f'{title}')
    print(f'  Authors: {', '.join(authors[:3])}')
    print(f'  ID: {link.split('/')[-1]}')
    print()
"

---

Search Tips

arXiv Categories

Category	Field
cs.AI	Artificial Intelligence
cs.CL	Computation and Language (NLP)
cs.LG	Machine Learning
cs.CV	Computer Vision
stat.ML	Machine Learning (Statistics)
cs.SE	Software Engineering
cs.CR	Cryptography and Security

Search Operators

Operator	Example	Effect
all:	all:transformer	Search all fields
ti:	ti:GRPO	Title only
au:	au:bengio	Author name
abs:	abs:benchmark	Abstract only
cat:	cat:cs.AI	Category filter
AND	all:RL AND cat:cs.LG	Both conditions
ANDNOT	all:transformer ANDNOT cat:cs.CL	Exclude category

---

Verification

# Check skill is installed
hermes skills list | grep arxiv

# Test API directly
curl -s "https://export.arxiv.org/api/query?search_query=all:test&max_results=1" | grep -c "<entry>"
# Expected: 1

---

Related Skills

• ocr-and-documents — Extract text from academic PDFs and scanned documents
• research-paper-writing — Write academic papers with proper formatting
• llm-wiki — Build a personal research wiki from collected papers

---

Pro Tips

1. Use sortBy=submittedDate — Always sort by date for literature reviews. Default is relevance, which biases toward older highly-cited papers.
2. Limit max_results — arXiv's API is rate-limited. Start with max_results=5 and increase only if needed.
3. Cache results — For literature reviews, save the XML output to a file and process locally rather than re-querying.
4. Read PDFs via web_extract — After finding a paper, use web_extract(urls=["https://arxiv.org/pdf/PAPER_ID"]) to get the full text without downloading.
5. Combine with humanizer — Use the humanizer skill to convert academic abstracts into plain-language summaries.