OMNI Research Verifier

Overview

This skill provides a high-assurance verification layer for agentic workflows. It is designed to be the "Truth Node" in a multi-agent system. Unlike standard search skills, OMNI Research Verifier doesn't just find sources — it evaluates them, scores them, and identifies contradictions.

Problem It Solves

Large language models are prone to hallucinations and fabricated citations. This skill forces the agent to step outside its internal weights and verify every claim against a live, multi-source ground truth.

How It Works

1. Claim Extraction

The skill deconstructs user input into individual claims that need verification.

2. Multi-Source Search

It queries across web and news sources simultaneously using the duckduckgo_search library.

3. Credibility Scoring

Each source is evaluated based on domain authority (.edu, .gov, .org get higher scores) and content quality.

4. Contradiction Detection

The skill scans for conflicting information across sources and flags them for user review.

5. Structured Report

Outputs a JSON report with:

• Executive Summary
• Evidence Table (source, credibility score, excerpt)
• Contradictions (if any)
• Formatted Citations
• Confidence Score (0-100)

Installation

claw install omni-research-verifier

Usage

from omni_research_verifier import process

result = process("The Earth's core has stopped rotating.")
print(result)

Example Output

{
  "status": "success",
  "confidence_score": 85,
  "executive_summary": "Claim is Mixed based on 5 sources.",
  "contradictions": [
    "Potential conflict identified in https://example.com"
  ],
  "evidence_table": [
    {
      "source": "https://example.com",
      "title": "Earth's core rotation study",
      "snippet": "The Earth's core has slowed...",
      "score": 0.7,
      "timestamp": "2026-08-18"
    }
  ],
  "citations": [
    "Earth's core rotation study. (2026-08-18). Retrieved from https://example.com"
  ]
}

Technical Requirements

· Python 3.10+
· Internet connection for web search
· OpenClaw agent environment

Security Notes

· No API keys required (uses duckduckgo_search which is free and anonymous)
· No persistent storage of search results
· All searches are performed anonymously via DuckDuckGo

License

MIT-0 — Free to use, modify, and redistribute. No attribution required.

Support

For issues or suggestions, please open an issue on GitHub or contact the maintainer.

Changelog

Version 1.0.0 (2026-08-18)

· Initial release
· Multi-source search capability
· Credibility scoring engine
· Contradiction detection
· Structured JSON output