building-data-apps
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of a chat interface that processes user-supplied text to interact with the Gemini Data Analytics API. This creates a functional surface for indirect prompt injection, which is a standard characteristic of LLM-integrated applications. The implementation includes system instructions to bound model behavior. Ingestion points: user messages and history are processed in examples/react_chat_panel.jsx and the backend endpoints. Boundary markers: backend examples define a system_instruction in inline_context to guide the model. Capability inventory: the model is instructed to generate SQL for BigQuery analysis. Sanitization: the code relies on model guardrails and specific instructions rather than explicit input sanitization.
- [EXTERNAL_DOWNLOADS]: The skill instructions and examples recommend standard, reputable libraries for development. These include official Google Cloud packages (@google-cloud/geminidataanalytics, @google-cloud/bigquery) and established community libraries such as streamlit, plotly, echarts, lucide-react, and react-markdown. All dependencies are consistent with the skill's intended purpose.
- [COMMAND_EXECUTION]: The provided scripts for FastAPI and Express use standard web server patterns. Analysis found no use of dangerous shell commands, insecure process spawning, or arbitrary execution of untrusted input.
Audit Metadata