firebase-vertex-ai
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform standard Firebase development tasks, such as project initialization (
firebase init), building TypeScript functions (npm run build), and deploying services to Google Cloud Platform. These operations are restricted to the local environment and the user's authorized projects. - [EXTERNAL_DOWNLOADS]: The
init-firebase.shscript facilitates the installation of the officialfirebase-toolspackage from the npm registry. This is a well-known, trusted utility required for the skill's primary functionality. All other documentation links point to official Google Cloud or Firebase domains. - [INDIRECT_PROMPT_INJECTION]: The skill includes templates for building AI features that process user-supplied messages or document content. While this creates a surface for indirect prompt injection, the skill mitigates risk by providing guidance on input validation, structured responses, and secure backend execution via Cloud Functions.
- Ingestion points: User-provided strings in chat endpoints (
req.data.message) and document content in Firestore triggers. - Boundary markers: The implementation examples use standard template literal interpolation for prompt assembly.
- Capability inventory: The generated code utilizes the Firebase Admin SDK for database interactions and the Vertex AI SDK for model inference.
- Sanitization: The skill emphasizes the use of LLM safety settings and structured JSON output to ensure model responses remain within expected parameters.
Audit Metadata