Claw Score

What This Skill Does

The Claw Score skill is a specialized architecture auditing tool designed for OpenClaw agents. It functions as a security and efficiency scanner that evaluates the structural integrity and quality of your agent's core instruction files. By analyzing key configuration files—such as AGENTS.md, SOUL.md, and SECURITY.md—the skill generates a professional-grade audit report. It automatically performs sensitive data sanitization, ensuring that credentials, PII, and environment variables are stripped before any transmission to the Atlas auditing server occurs. You receive a structured feedback report within 24-48 hours, detailing your agent's performance across six distinct dimensions, complete with actionable recommendations for optimization.

Installation

To integrate the Claw Score audit capabilities into your current environment, you can use the OpenClaw skill management system. The primary method is via the command line interface provided by ClawHub. Execute the following command in your agent's root workspace:

npx clawhub install claw-score

Alternatively, if you are managing your skills manually, ensure the claw-score folder is correctly mapped to your local skills/ directory. Once the files are in place, the agent will recognize the new commands during the next reload of the skill registry. No secondary dependencies are required for the local processing of file analysis.

Use Cases

• Agent Optimization: Use this tool during the development phase to ensure your system prompts and tool definitions follow best practices for logic and clarity.
• Security Hardening: Before deploying an agent to production, audit your SECURITY.md and IDENTITY.md files to ensure no leaky instructions could compromise your agent's intent.
• Compliance Checks: Periodically audit your workspace to ensure that your agent's long-term memory config and proactive behavior settings align with your organizational standards.
• Performance Benchmarking: Obtain a 'Claw Score' to track the maturation of your agent's persona and logic over time as you iterate on its instruction files.

Example Prompts

1. "Run a Claw Score audit and send the report to [email protected]"
2. "Audit my current agent architecture and provide recommendations for improvement. Use [email protected] for the delivery."
3. "Submit my workspace for a full Claw Score audit now. My contact email is [email protected]"

Tips & Limitations

• Sanitization Transparency: Always review the generated preview before confirming the audit submission to ensure the sanitization process hasn't accidentally altered necessary structural instructions.
• File Requirements: The audit is most effective when your workspace contains the full suite of suggested files (e.g., SOUL.md, MEMORY.md, TOOLS.md). If files are missing, the audit report will be less comprehensive.
• Wait Times: Remember that the analysis is performed by the Atlas service, which operates on a 24-48 hour turnaround. If you require immediate feedback, check the audit-framework.md documentation for manual scoring criteria.
• No Code Execution: This skill strictly reads .md files. It does not run scripts or execute code, making it inherently safer than traditional diagnostic tools.