arXiv Viewer

Access, search, download, and read academic papers from arXiv using a hybrid API + Actionbook approach.

Architecture Overview

┌─────────────────────────────────────────────────────────────┐
│                     arxiv-viewer                            │
├─────────────────┬─────────────────┬─────────────────────────┤
│   arXiv API     │  arxiv.org Web  │      ar5iv.org          │
│   (WebFetch)    │  (Actionbook)   │    (Actionbook)         │
├─────────────────┼─────────────────┼─────────────────────────┤
│ • Metadata      │ • Latest list   │ • Read sections         │
│ • Search        │ • Trending      │ • Extract figures       │
│ • By ID lookup  │ • Advanced      │ • Extract citations     │
│                 │   search form   │ • Get outline           │
└─────────────────┴─────────────────┴─────────────────────────┘

Tool Priority

By Feature:

Workflow Rules

⚠️ Agent Waiting Rule

After launching browser-fetcher or html-reader agents:

1. ✅ MUST wait for ALL agents to complete
2. ⛔ DO NOT use WebFetch/WebSearch while waiting
3. ✅ Only use fallback tools after agents have failed

Data Sources

1. arXiv API (WebFetch)

Best for: Quick metadata lookup, simple search

Base URL: http://export.arxiv.org/api/query

Field Prefixes: ti: (title), au: (author), abs: (abstract), cat: (category), all: (all)

Boolean Operators: AND, OR, ANDNOT (UPPERCASE)

2. arxiv.org Web (Actionbook + agent-browser)

Best for: Latest papers, trending, advanced search UI

Workflow:

1. search_actions("arxiv list recent")
2. get_action_by_id(action_id) → selectors
3. agent-browser open URL
4. agent-browser get text <selector>
5. Return results

3. ar5iv.org HTML Papers (Actionbook + agent-browser)

Best for: Reading specific sections, extracting figures/citations

HTML Paper URL: https://ar5iv.org/html/{arxiv_id}

Workflow:

1. search_actions("ar5iv section")
2. get_action_by_id(action_id) → selectors
3. agent-browser open ar5iv.org/html/{id}
4. agent-browser get text <section_selector>
5. Return section content

URL Patterns

arXiv ID Formats

Common Categories

Feature Matrix

Output Formats

Paper Info

## {Title}

**arXiv:** {id}
**Authors:** {author1}, {author2}, ...
**Categories:** {cat1}, {cat2}
**Published:** {date}

### Abstract
{abstract}

**Links:** [Abstract]({abs_url}) | [PDF]({pdf_url}) | [HTML]({ar5iv_url})

Section Content

## {Section Title}

{section content}

---
*Source: ar5iv.org/html/{id}*

Paper Report (AI Generated)

Command: /arxiv-viewer:report {arxiv_id} or /arxiv-viewer:report {paper_title}

Purpose: Generate a comprehensive, well-formatted paper report with AI-generated analysis.

Output Format:

---
> **🤖 AI Generated Content**
> Author: Powered by ActionBook
---

# {Paper Title}

**Paper Information**
| Field | Content |
|-------|---------|
| arXiv ID | {id} |
| Authors | {authors} |
| Affiliations | {affiliations} |
| Published | {date} |
| Categories | {categories} |

---

## 📋 Abstract

{abstract_summary}

---

## 🎯 Problem Statement

{problem_statement}

---

## 💡 Key Contributions

1. {contribution_1}
2. {contribution_2}
3. {contribution_3}

---

## 🔬 Method Overview

{method_summary}

---

## 📊 Experimental Results

{experimental_results}

---

## 🌟 Why It Matters

{significance}

---

## 🔗 Links

- [arXiv Abstract](https://arxiv.org/abs/{id})
- [PDF Download](https://arxiv.org/pdf/{id}.pdf)
- [HTML Version](https://ar5iv.org/html/{id})

---
> This report was automatically generated by AI based on the original paper content.
> **Powered by ActionBook** | Generated: {timestamp}

Workflow:

1. Fetch paper metadata via arXiv API
2. Read full paper content from ar5iv.org HTML
3. Extract key sections (abstract, intro, method, results)
4. Generate structured report with AI analysis
5. Add ActionBook branding header and footer

Report Sections:

Rate Limiting

• arXiv API: 3 second delay between requests
• ar5iv.org: Respect server load
• agent-browser: One page at a time