A self-correction protocol that prevents confident AI errors by verifying claims before delivery.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install self-verify
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 self-verify 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 Self-Verify?
Self-Verify is a critical reliability layer for AI agents, designed to act as a final gatekeeper before a response is finalized. It implements a rigorous no confident nonsense policy by mandating evidence-based verification for consequential answers, ensuring that every high-stakes claim is cross-referenced against search data or established memory. By integrating this into Openclaw Skills, developers can drastically reduce hallucinations and improve the technical integrity of agentic workflows.
This skill is particularly essential for agents operating at high velocity where the risk of confident but incorrect assertions is high. It forces a pause in the reasoning process to apply a standardized checklist, ensuring that uncertainty is labeled and contradictions are identified before the user ever sees the output.
Self-Verify Use Cases
- Before providing quantitative facts or data-driven conclusions.
- When an agent is asked to confirm if a specific claim is true.
- Prior to giving actionable advice that involves high stakes or irreversibility.
- When a user explicitly questions the agent with phrases like "are you sure" or "verify this".
- Before contradicting information previously stored in long-term memory.
How Self-Verify Works
- The agent identifies a high-consequence claim or a specific user trigger for verification during the reasoning phase.
- It executes a comprehensive 5-point checklist covering source credibility, uncertainty levels, internal contradictions, reversibility of advice, and safety risks.
- The skill triggers external tools such as mmx search or internal memory search to validate information against external facts or historical context.
- Logic chains are analyzed step-by-step to identify speculative leaps or steps with weak evidence.
- The final output is formatted with explicit confidence levels and evidence citations to provide full transparency to the user.
Self-Verify Setup
To integrate Self-Verify into your environment, ensure your agent has access to search and memory tools. Add the skill configuration to your Openclaw Skills directory.
# Install the self-verify protocol
mcp install self-verify
# Ensure search and memory tools are configured for cross-checking
mmx setup-search
memory_search --init
Self-Verify Data Schema & Taxonomy
The skill organizes verification data using a structured taxonomy to distinguish between observed facts and inferences.
| Label | Description |
|---|---|
| OBS | Observed fact from a credible source |
| DER | Derived information based on logic |
| INT | Internal agent interpretation |
| SPEC | Speculative or high-uncertainty claim |
Verified outputs are returned using a standardized structure:
- Self-verified: Concise claim
- Evidence: Source or reasoning chain
- Confidence: HIGH/MEDIUM/LOW
- Uncertainty: Explicit identification of what is not known
Self-Verify Advanced Features
- Real-time cross-checking using mmx search for live web verification of claims.
- Memory.md integration to ensure consistency with historical session data and established facts.
- Automated reasoning trace analysis that flags specific logic steps as SPEC or LOW CONFIDENCE.
- Heuristic triggers for linguistic patterns like "definitely" or "the only solution" to prevent false precision.
- Safety-first HOLD verdicts that require human review for irreversible or high-risk claims.
SKILL.md
Loading
Related Openclaw Skills
Bangboo Voice Agent Skill
A cross-platform persona skill that enables AI agents to roleplay as Zenless Zone Zero Bangboos while maintaining technical accuracy and safety.
AI Video Generator Free HeyGen
A powerful AI tool that converts text scripts into professional-grade talking avatar videos without requiring complex editing software or expensive subscriptions.
Spanish Income Tax Assistant (IRPF 2025)
A professional-grade assistant designed to help Spanish taxpayers review, optimize, and prepare their 2025 Income Tax (IRPF) returns using national and regional regulations.
Pipeworx Lorem
A high-performance Model Context Protocol (MCP) server that provides instant access to realistic placeholder text via the loripsum.net API.
PV-52 Crisis Consultant
A high-pressure risk management AI agent designed to provide structural stability and strategic decision-making during organizational crises.
JobHound
JobHound is a comprehensive AI-driven career companion designed to guide candidates through the entire recruitment lifecycle, from initial job discovery to final salary negotiation.
Featured*
