ai-safety-guard

What This Skill Does

The AI Safety Guard is a foundational behavioral skill designed to transform OpenClaw into a privacy-first assistant. Unlike traditional filtering tools that run externally, this skill integrates directly into the AI's core processing logic. It ensures that sensitive information—including government-issued IDs, bank credentials, phone numbers, medical records, and passwords—is automatically detected and sanitized before it is rendered in any output. By embedding privacy awareness into the AI's decision-making flow, it prevents accidental leakage during document drafting, email composition, and conversational memory recall. It provides multiple levels of protection, from silent suppression to interactive confirmation, ensuring users retain control over how their data is handled.

Installation

To activate the AI Safety Guard, run the following command in your OpenClaw terminal: clawhub install openclaw/skills/skills/andreqingyuwu/ai-safety-guard After installation, the AI will automatically begin scanning outputs for sensitive patterns. No manual configuration is required for the default silent filter mode.

Use Cases

This skill is essential for users handling sensitive workflows, such as:

• Professional Communications: Drafting emails or reports that reference sensitive client data or internal company credentials.
• Content Management: Exporting chat history or documents where personal identifiers, addresses, or financial data might be embedded in the text.
• Credential Handling: Working with API keys or tokens where the AI needs to assist in development without exposing raw secrets to the terminal buffer or shared screen.
• Data Privacy Compliance: Ensuring that AI-generated summaries of conversations remain compliant with internal privacy policies by stripping out PII before storage or sharing.

Example Prompts

1. "Draft an email to the insurance provider attaching my medical report details but ensure all patient IDs are redacted."
2. "Summarize the previous conversation regarding the project budget and export the text, ensuring my bank account numbers are removed."
3. "Can you help me write a follow-up email to the landlord including my current home address while keeping my private social security number hidden?"

Tips & Limitations

• Proactive Vigilance: The AI activates automatically, but you can set the protection level (1-4) in your settings to match your risk profile.
• False Positives: If the AI incorrectly flags non-sensitive data, providing feedback helps improve the model's recognition accuracy over time.
• Boundary Limitations: While highly effective, this skill relies on pattern recognition; always exercise caution when handling hyper-sensitive cryptographic keys or unique physical assets.