open_research_skill

Multi-source open-access research for AI agents — with citation provenance that prevents hallucinated references.

---

What is it?

open_research_skill is a research automation tool for AI agents that searches across five open-access academic sources, reads full papers, and enforces a citation provenance system that prevents hallucinated references.

Every citation is grounded in actual document content — no more fabricated papers, wrong DOIs, or made-up quotes.

Why it exists

AI systems generate plausible-sounding but entirely fabricated references at alarming rates:

• ChatGPT's references exist only 14% of the time [Zuccon et al., 2023]
• Bard hallucinated 91.4% of systematic review references [Chelli et al., 2024]

This tool solves that problem by requiring every cited claim to be verified against actual document content.

Features

🔍 Multi-source search

Source	Coverage	Full Text	API Key
OpenAlex ⭐	250M+ works	Selective (OA)	No
arXiv	2.3M+ papers	Yes (PDF)	No
CrossRef	150M+ DOIs	Metadata	No
Semantic Scholar	200M+ papers	Selective	Optional
PubMed Central	8M+ articles	Yes (XML)	No

📄 Four-tier citation provenance

Tier	What was read	Can cite?
Read	Full PDF, page-level	✅ Yes, with page numbers
Abstract	Abstract text	✅ Yes, abstract claims only
Metadata	Title, authors, year, DOI	⚠️ Context only
Discovery	Search result only	❌ No

✅ Claim verification

• Download PDFs and extract text with page numbers
• Search for specific claims and return page-level evidence
• Generate bibliography verification files for human review

🧹 Deduplication

• Identifies the same paper across multiple sources
• Priority: DOI → arXiv ID → PMC ID → title similarity

Quick Start

# Install dependencies
pip install arxiv crossref semanticscholar requests PyMuPDF pdfplumber

# Search across all sources
python research_workflow.py search "porphyria transcriptomics" --max-results 5

# Read a full paper
python research_workflow.py read-paper 10.5281/zenodo.6936227

# Full analysis with paper reading
python research_workflow.py full-analysis "AI research agents" --read-papers

# Generate bibliography verification
python research_workflow.py bibliography --output bib_verify.json

Programmatic usage

from research_workflow import (
    search_openalex, search_arxiv, search_crossref,
    search_pmc, search_semantic_scholar,
    read_paper_full, verify_claim_in_paper,
    generate_bibliography
)

# Search
papers = search_openalex("gene therapy", max_results=10)

# Read a paper with provenance tracking
citation = read_paper_full(
    doi="10.1016/j.example.2024.01.001",
    title="Example Paper Title",
    authors="Author, A.",
    year=2024,
    source="OpenAlex"
)

# Verify a claim in a previously-read paper
verified, locations = verify_claim_in_paper("paper_20260512_120000", "the treatment improved outcomes by 40%")

Project Structure

openresearch/
├── research_workflow.py          # Main workflow (897 lines)
├── CITATION.cff                  # Machine-readable citation metadata
├── .zenodo.json                  # Zenodo archive metadata
├── test_research_tools.py        # Test suite
├── references.bib                # Bibliography (BibTeX)
├── research_paper.tex       # Research paper (LaTeX)
├── papers/                       # Downloaded PDFs
├── pages/                        # Page-level extracted text
└── *.json                        # Search results

Citation Provenance in Action

{
  "identifiers": {"doi": "10.5281/zenodo.6936227"},
  "title": "OpenAlex: A fully-open index of scholarly works...",
  "provenance": {
    "tier": "read",
    "page": 3,
    "section": "Introduction",
    "claim_verified": true,
    "abstract_text": "OpenAlex is a fully-open index..."
  },
  "verified_claims": [
    {
      "claim": "OpenAlex contains 250M+ works",
      "verified": true,
      "page": 3,
      "context": "...OpenAlex provides a comprehensive index of over 250 million scholarly works..."
    }
  ]
}

Limitations

• Open access only — paywalled papers are identified but cannot be read
• PDF parsing — works well for most PDFs, may struggle with scanned documents
• Semantic Scholar — free tier is rate-limited; API key recommended for production use

License

MIT License — feel free to use, modify, and contribute.

BibTeX

@software{ruizpalmero2026open_research_skill,
  title = {{open\_research\_skill: Multi-Source Open-Access Research for AI Agents}},
  author = {Ruiz-Palmero, Christian},
  year = {2026},
  version = {1.0.0},
  url = {https://github.com/ruizcrp/open_research_skill},
  license = {MIT}
}

Related Work

• Orchestra-Research/AI-Research-SKILLs — Comprehensive AI research skills library
• Feynman — AI research agent that reads papers and writes drafts
• AI Scientist — Autonomous scientific discovery with LLMs
• LitLLM — Literature review toolkit for LLMs

Contributing

Contributions are welcome! Whether it's bug fixes, new search sources, or documentation improvements — just open an issue or PR.

Citation

If you use this tool in your research, please cite it:

@software{ruizpalmero2026open_research_skill,
  title = {{open\_research\_skill: Multi-Source Open-Access Research for AI Agents}},
  author = {Ruiz-Palmero, Christian},
  year = {2026},
  version = {1.0.0},
  url = {https://github.com/ruizcrp/open_research_skill},
  license = {MIT}
}