recipe-setup-sanitization

Set Up Global Response Sanitization

Configure Google Cloud Model Armor to scan all gws API responses for prompt injection before they reach AI agents.

When to Use

Use this workflow when the user wants to set up response sanitization — either globally for all commands or for specific high-risk operations like reading untrusted emails.

Prerequisites

• GCP project with Model Armor API enabled
• Full scopes: gws auth login --full

Steps

1. Create a Model Armor template:

   gws modelarmor +create-template \
     --project my-project-id \
     --location us-central1 \
     --template my-safety-template
   

2. Test sanitization on a single command:

   gws gmail users messages get \
     --params '{"userId": "me", "id": "MSG_ID"}' \
     --sanitize "projects/my-project-id/locations/us-central1/templates/my-safety-template"
   

3. Enable globally via environment variables:

   # Add to your .env or shell profile
   export GOOGLE_WORKSPACE_CLI_SANITIZE_TEMPLATE="projects/my-project-id/locations/us-central1/templates/my-safety-template"
   export GOOGLE_WORKSPACE_CLI_SANITIZE_MODE=warn
   

4. Choose a mode:

   Mode	Behavior
   warn (default)	Logs a warning but returns the response
   block	Returns an error if injection is detected

5. Verify it's active:

   # Any gws command will now show sanitization status in debug logs
   GOOGLE_WORKSPACE_CLI_LOG=gws=debug gws gmail +triage
   

6. Make it permanent — add to .env file:

   cat >> .env << 'EOF'
   GOOGLE_WORKSPACE_CLI_SANITIZE_TEMPLATE=projects/my-project-id/locations/us-central1/templates/my-safety-template
   GOOGLE_WORKSPACE_CLI_SANITIZE_MODE=warn
   EOF
   

Per-Command Override

Even with global sanitization, you can override per-command:

# Use a different template for sensitive operations
gws gmail users messages get \
  --params '{"userId": "me", "id": "MSG_ID"}' \
  --sanitize "projects/P/locations/L/templates/strict-template"