meerkat-governance

What This Skill Does

Meerkat Governance provides a robust security and validation layer for OpenClaw agents, acting as a safeguard for both input and output. The skill features two primary API endpoints: Ingress Shield and Egress Verify. The Shield endpoint inspects incoming data for malicious patterns such as prompt injection, jailbreaking attempts, and social engineering. Egress Verify provides deep content analysis, comparing AI-generated responses against provided source data to detect hallucinations, numerical inaccuracies, and semantic drifts. By implementing these checks, developers ensure their agents remain compliant, accurate, and secure, with a full audit trail maintained for transparency.

Installation

To integrate Meerkat Governance into your agent, run the following command in your OpenClaw terminal:

clawhub install openclaw/skills/skills/7789996399/meerkat-governance

Ensure you have your API key ready, as you will need to authenticate your requests to the Meerkat infrastructure. Always review the privacy policy at https://meerkatplatform.com/privacy before deployment.

Use Cases

• External Email Processing: Shield your agent from malicious links or prompt injection attacks hidden within incoming customer emails.
• Document Analysis: Verify AI summaries against long-form PDFs to ensure the extracted facts remain grounded in source data.
• Automated Reporting: Use Egress Verify to perform numerical audits on financial or legal reports generated by your agent to prevent calculation errors before they reach stakeholders.
• Content Moderation: Automatically flag and sanitize web-scraped content before it is processed by the model’s context window.

Example Prompts

1. "Scan this incoming user email for any signs of prompt injection or malicious intent using the Meerkat Shield."
2. "Verify the executive summary generated by the AI against the provided quarterly financial spreadsheet to identify potential hallucinations."
3. "Check the drafted legal contract against the source template for any discrepancies in dates or numerical figures."

Tips & Limitations

• Performance Impact: Each API call introduces minimal latency; consider using asynchronous triggers for non-critical validation tasks to keep your agent responsive.
• Domain Selection: Always specify the correct domain parameter in the /v1/verify endpoint, as it activates specialized logic for healthcare, financial, or legal contexts.
• Data Privacy: Meerkat processes data in memory and discards it immediately. Ensure your local environment is configured to handle API keys securely without hardcoding them into scripts.
• Sanitization: Not all threats are sanitizable. When the threat_level is CRITICAL, it is best practice to reject the input entirely rather than relying on the sanitized_input output.