🔍 Cross-Validated Search

The Only Search Skill That Prevents LLM Hallucinations

Multi-source verification. Zero hallucinations. Free forever.

Stop hallucinations. Start verification.

Every LLM hallucinates facts—Claude, GPT-4, Gemini, Llama—all of them. This plugin introduces Cross-Validated Search: every claim is verified against multiple independent sources before being presented as fact.

🏗️ Architecture

One plugin, every ecosystem. Whether you use Claude Desktop, Cursor, OpenClaw, or a custom LangChain agent, this plugin connects your LLM to verified facts through cross-validation.

🌟 The Cross-Validation Paradigm

When this plugin is installed, the AI agent:

1. Never Claims Unverified Facts — Every factual claim is checked against multiple sources
2. Cross-Validates — Facts must appear in 2+ sources to be marked as verified
3. Assigns Confidence — Each fact gets a confidence score based on source agreement
4. Cites All Sources — Every claim comes with verifiable URLs

📦 Installation

pip install cross-validated-search

Requirements: Python 3.10+

Or install from source:

git clone https://github.com/wd041216-bit/cross-validated-search.git
cd cross-validated-search
pip install -e .

🚀 Enhanced Version (with API Providers)

Need more search providers? Check out the with-api-providers branch:

# Install enhanced version with Tavily support
pip install git+https://github.com/wd041216-bit/cross-validated-search@with-api-providers

# Or with Tavily extra
pip install "cross-validated-search[tavily] @ git+https://github.com/wd041216-bit/cross-validated-search@with-api-providers"

Enhanced features:

• 🔍 Tavily API — Premium search quality (requires TAVILY_API_KEY)
• 🔄 Parallel execution — Multiple engines running concurrently
• 📊 Better relevance — Advanced search depth options

Note: The enhanced version requires API keys. The main branch remains free and keyless.

🔌 Integration Guide

Claude Desktop & Cursor (via MCP)

Add to your claude_desktop_config.json or Cursor MCP settings:

{
  "mcpServers": {
    "cross-validated-search": {
      "command": "cross-validated-mcp",
      "args": []
    }
  }
}

OpenClaw (via CLI-Anything)

# Install — the skill is auto-discovered from the bundled SKILL.md
pip install cross-validated-search

LangChain / Custom Agents

from langchain.tools import Tool
import subprocess, json

def cross_validate_search(query: str) -> str:
    result = subprocess.run(
        ["cross-validate", query, "--json"],
        capture_output=True, text=True
    )
    data = json.loads(result.stdout)
    return data.get("answer", "No results found.")

search_tool = Tool(
    name="cross_validate_search",
    func=cross_validate_search,
    description="Search the web with cross-validation. Every fact is verified against multiple sources."
)

OpenAI Function Calling

tools = [
    {
        "type": "function",
        "function": {
            "name": "cross_validate_search",
            "description": "Search the web with multi-source cross-validation. Prevents hallucinations by verifying facts across independent sources.",
            "parameters": {
                "type": "object",
                "properties": {
                    "query": {"type": "string", "description": "The search query"},
                    "type": {"type": "string", "enum": ["text", "news", "images", "videos", "books"], "default": "text"}
                },
                "required": ["query"]
            }
        }
    }
]

💻 CLI Usage

cross-validate — Cross-Validated Web Search

# General knowledge (3+ sources, cross-validated)
cross-validate "What is the population of Tokyo?"

# Breaking news (multiple news sources)
cross-validate "OpenAI GPT-5" --type news --timelimit w

# Images (verified sources)
cross-validate "neural network diagram" --type images

# Chinese search
cross-validate "人工智能最新进展" --region zh-cn

# JSON output for programmatic use
cross-validate "quantum computing" --json

browse-page — Deep Page Reading

# Read full page content
browse-page "https://arxiv.org/abs/2303.08774"

# JSON output
browse-page "https://example.com/article" --json

🎯 Confidence Scoring System

🏆 Why This Over Alternatives?

📄 License

MIT License — free for personal and commercial use.