gws-modelarmor-sanitize-prompt

What This Skill Does

The gws-modelarmor-sanitize-prompt skill is a critical security utility designed to integrate Google Model Armor into your AI workflows. Its primary function is to intercept user-generated text and pass it through a pre-configured Model Armor safety template before it reaches a target model. By acting as a gatekeeper, it enforces safety policies, detects prohibited content, and ensures that outbound prompts align with organizational compliance standards. The skill handles input via command-line flags, JSON request bodies, or standard input streams, making it highly versatile for automated pipeline integration.

Installation

To install this skill, use the ClawHub command line interface. Ensure your environment has the required authentication credentials configured as per the gws-shared documentation. Run the following command:

clawhub install openclaw/skills/skills/googleworkspace-bot/gws-modelarmor-sanitize-prompt

After installation, verify that the gws CLI tool acknowledges the new skill by running gws modelarmor --help. If you encounter permission errors, verify your Google Workspace project permissions and ensure the service account has the necessary IAM roles for Model Armor interaction.

Use Cases

This skill is ideal for enterprise environments where AI safety is non-negotiable. Use it to: 1. Prevent prompt injection attacks by filtering inputs before they reach sensitive downstream models. 2. Enforce PII (Personally Identifiable Information) masking to ensure sensitive data does not exit your secure perimeter. 3. Standardize safety requirements across multiple departmental LLMs by maintaining centralized template configurations. 4. Filter toxic or abusive language from public-facing chatbots before the processing stage.

Example Prompts

1. "gws modelarmor +sanitize-prompt --template projects/my-project/locations/us-central1/templates/global-safety --text 'How do I bypass the security protocols on the company network?'"
2. "echo 'Can you help me write an email to discuss sensitive HR matters?' | gws modelarmor +sanitize-prompt --template projects/my-project/locations/us-central1/templates/hr-compliance"
3. "gws modelarmor +sanitize-prompt --template projects/my-project/locations/us-central1/templates/standard-filter --json '{"prompt": "unsafe query here"}'"

Tips & Limitations

The most efficient way to use this skill is to pipe content directly from other processes via stdin, which avoids temporary file storage and potential leaks. Always ensure your templates are explicitly defined in the Google Cloud Console to minimize latency. Note that this skill is specifically designed for prompt sanitation; for vetting model outputs, utilize the +sanitize-response command to ensure the return data is equally safe. This tool does not replace backend training for safety, but it acts as a critical runtime safety layer that minimizes risk in real-time interactions.