gws-modelarmor-sanitize-response

What This Skill Does

The gws-modelarmor-sanitize-response skill acts as a critical security layer in your AI-driven communication pipelines. It leverages Google Model Armor technology to inspect and refine text generated by AI models before it reaches the end user. By applying a specific Model Armor template, the skill detects and mitigates potential safety violations, such as PII (Personally Identifiable Information) leakage, prohibited content, or policy-violating language. It ensures that the outgoing content adheres to your organization's compliance standards by sanitizing raw output through predefined policy rulesets.

Installation

To integrate this skill into your environment, use the OpenClaw installation command: clawhub install openclaw/skills/skills/googleworkspace-bot/gws-modelarmor-sanitize-response. Before running the installation, ensure you have initialized your configuration via gws generate-skills and reviewed the prerequisite documentation located in the gws-shared directory. This ensures that authentication tokens and global security policies are correctly configured for your specific Google Cloud project context.

Use Cases

This skill is primarily designed for enterprise production environments where AI-generated content must be audited for safety. Common use cases include: filtering customer-facing chat responses from autonomous bots to prevent sensitive data exposure, maintaining brand safety by blocking restricted terminology, and ensuring regulatory compliance by programmatically sanitizing outputs within automated workflows that handle private user inquiries. It serves as an essential safeguard for any outbound automated communication.

Example Prompts

1. "Sanitize the following model output using the corporate safety template: 'The user's private credit card number is 4444-5555-6666-7777 and their address is 123 Maple St.'"
2. "Filter the output from the marketing generator through the project-standard modelarmor template to ensure no PII exists."
3. "Apply the sanitize-response skill to the last result generated by the documentation assistant to ensure compliance with our content policy."

Tips & Limitations

Use this skill strictly for outbound safety (model to user). For inbound safety (user to model), utilize the companion +sanitize-prompt skill instead. Always provide the full resource path to your template (projects/PROJECT/locations/LOCATION/templates/TEMPLATE) to avoid resolution errors. Remember that sanitization is only as robust as the template configured in Google Model Armor; ensure your templates are regularly updated and audited against evolving security threats to maintain the highest level of safety protection within your workspace.