GCP Agent Safety Gatekeeper

gcp-agent-safety-gatekeeper

This skill implements the Python integration layer for Model Armor. Grounded in security_blog.md, it provides the safety_util functions needed to intercept prompts, sanitize them against your security policy, and handle safety triggers in your FastAPI backend.

Usage

Ask Antigravity to:

• "Add a safety gatekeeper to my agent backend"
• "Implement Model Armor prompt sanitization in Python"
• "Create a safety utility to parse Model Armor findings"
• "Handle prompt injection errors in my FastAPI app"

Integration Pattern

1. Client Initialization: Configures the ModelArmorClient with the correct regional endpoint.
2. safety_util.py: A robust parser that converts SanitizeUserPromptResponse into a list of human-readable security triggers (e.g., "Prompt Injection", "PII: Person names").
3. Application Interception: Logic to block or sanitize prompts before they reach the GenAI model or agent orchestrator.

Boilerplate Implementation

Refer to scripts/safety_util.py for the core parsing logic.