ThoughtProof is a multi-agent verification protocol that ensures AI outputs are epistemically sound by triangulating reasoning across multiple independent model families.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install thoughtproof
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install thoughtproof using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is ThoughtProof?
ThoughtProof is a specialized verification skill designed for the ecosystem of Openclaw Skills. It functions as an epistemic auditor for AI-driven decision-making, ensuring that reasoning is not limited by the inherent biases or hallucinations of a single Large Language Model. By implementing a multi-agent consensus protocol, ThoughtProof allows developers to build trust in autonomous outputs by verifying claims against a diverse set of providers.
This skill operates on the fundamental principle that no single provider should verify itself. It enforces a strict model diversity requirement, mandating that at least three distinct model families—such as Anthropic, OpenAI, xAI, or Moonshot—participate in the reasoning process. The result is a synthesized consensus that highlights contradictions, identifies risks, and provides a measurable confidence score for any given AI output.
ThoughtProof Use Cases
- Validating high-stakes business, legal, or technical decisions where a single model might have blind spots.
- Generating a verifiable audit trail for regulatory compliance and governance in automated workflows.
- Detecting and mitigating provider-specific bias by triangulating reasoning across diverse model families.
- Performing deep-dive analysis on complex investment theses or cross-domain questions using rotated agent roles.
How ThoughtProof Works
- Normalization: The input question or claim is normalized to ensure consistent interpretation across different backend model families.
- Generation: The skill prompts at least three independent models to provide independent reasoning and proposals.
- Critique: An adversarial critic layer identifies potential blind spots, contradictions, and risks within the initial proposals.
- Evaluation: The system evaluates the critiques to determine the weight of various arguments.
- Synthesis: A final synthesizer produces an Epistemic Block containing the consensus score, confidence level, and any dissenting viewpoints.
ThoughtProof Setup
To begin using ThoughtProof with your Openclaw Skills setup, first install the required CLI tool:
npm install -g pot-cli
Next, configure your model providers in your ~/.potrc.json file. Ensure you have at least three different model families configured to satisfy the diversity protocol:
{
"generators": [
{ "provider": "xai", "model": "grok-4-1-fast" },
{ "provider": "moonshot", "model": "kimi-k2.5" },
{ "provider": "anthropic", "model": "claude-sonnet-4-6" }
],
"critic": { "provider": "anthropic", "model": "claude-opus-4-6" },
"synthesizer": { "provider": "anthropic", "model": "claude-opus-4-6" }
}
You can verify your configuration at any time by running tp config.
ThoughtProof Data Schema & Taxonomy
Every verification process generates an Epistemic Block, which is stored locally as a JSON object. This structure provides a transparent audit trail of the reasoning process.
| Component | Description |
|---|---|
| Proposals | Independent reasoning and answers from each generator model. |
| Critique | Adversarial analysis identifying logical gaps or provider bias. |
| Synthesis | The final result including consensus score and confidence level. |
| MDI | Model Diversity Index, measuring the statistical independence of the reasoning. |
Review your history of blocks using tp list and inspect specific results with tp show <n>.
ThoughtProof Advanced Features
- Contextual Chaining: Use the
--context lastflag to link current verifications to previous results, maintaining epistemic continuity. - Deep Verification: The
tp deepcommand runs multiple iterations with rotated model roles for meta-synthesis in high-stakes scenarios. - Model Diversity Index (MDI): A unique metric within this Openclaw Skills integration that quantifies the diversity and independence of the participating agents.
- Bring Your Own Key (BYOK): Full data sovereignty; all API calls are made directly from your environment to the model providers without intermediate routing.
SKILL.md
Loading
Related Openclaw Skills
Thor Image Generation Skill
A local image generation bridge for AI agents using the powerful Thor ComfyUI pipeline.
m365-pnp-cli
A comprehensive command-line toolset for managing Microsoft 365 services including SharePoint, Teams, and Entra ID through your AI agent.
grazy
A comprehensive command-line companion for real-time Graz city data, including public transport, weather, and events.
Web Deploy GitHub Pages
An autonomous workflow for building and deploying static sites to GitHub Pages using GitHub Actions.
X Expert
A conversation-driven AI assistant designed to streamline the research, drafting, and publishing process for X (Twitter) through a structured workflow.
Emperor Meeting: Multi-Agent Strategic Wisdom System
Emperor Meeting is a sophisticated Multi-Agent decision-making framework that leverages Maoist dialectics and historical leadership wisdom to solve complex strategic and organizational challenges.
Featured*
