grounding-with-google-genai
Pass
Audited by Gen Agent Trust Hub on Aug 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a bundled Python script,
scripts/google_genai_grounding.py, using theuv runcommand. This is a standard and expected use of local scripts to facilitate communication with the Google GenAI API. - [EXTERNAL_DOWNLOADS]: The skill fetches data from external HTTP/HTTPS URLs to provide grounding context. It includes a dedicated validation function,
_validate_url, which checks hostnames and IP addresses to prevent access to localhost or private network addresses, effectively mitigating SSRF (Server-Side Request Forgery) risks. - [PROMPT_INJECTION]: The skill processes untrusted content from user-provided URLs, which represents a surface for indirect prompt injection.
- Ingestion points: Untrusted data is retrieved from external websites via the
urlcommand inscripts/google_genai_grounding.py. - Boundary markers: The script uses a simple text header ('URLs to retrieve:') to separate the grounding context from the user prompt.
- Capability inventory: The skill is limited to performing stateless API calls and returning text; it does not possess capabilities to execute code, modify files, or maintain persistence based on the fetched data.
- Sanitization: The skill validates the destination of network requests (SSRF protection) and avoids hardcoding credentials, though it does not perform semantic filtering of retrieved content, relying on the underlying LLM's guardrails.
Audit Metadata