google-adk
Pass
Audited by Gen Agent Trust Hub on Sep 5, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill provides legitimate instructions for agent behavior, such as 'Always call get_order_status before answering an order question' and 'Never invent an order status'. These are appropriate boundaries for the specified use case and do not attempt to override safety filters or extract system prompts.
- [DATA_EXPOSURE]: The instructions explicitly recommend best practices for data security, such as keeping secrets and API keys out of prompts and injecting them at the tool layer instead. The code example uses mock data and does not access sensitive local files.
- [REMOTE_CODE_EXECUTION]: The skill references standard Google ADK and GenAI libraries. While it mentions the framework's built-in code execution tool, it does so in a documentation context and explicitly warns against allowing unvalidated input to reach sensitive tools like shell or SQL executors.
- [INDIRECT_PROMPT_INJECTION]: The skill addresses the surface area for indirect injection by recommending strict input validation for all tools and advising against relying on prompt text for access control enforcement, which are standard security mitigations.
Audit Metadata