custom-golden-image-discovery
Pass
Audited by Gen Agent Trust Hub on May 7, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches JSON configuration maps from
https://www.gstatic.com. This is an official, well-known Google domain for hosting static assets and is a trusted source in the context of a Google Cloud Platform skill.\n- [COMMAND_EXECUTION]: The agent is instructed to usecurlto retrieve mapping data. The command dynamically constructs a URL path using a version string extracted from user input. This represents a safe and expected operational pattern for the skill's discovery purpose.\n- [PROMPT_INJECTION]: The skill processes untrusted user data to construct the parameters for a network request, creating a surface for indirect prompt injection.\n - Ingestion points: User-provided setup descriptions or GKE version strings parsed during the 'Discovery Workflow' in SKILL.md.\n
- Boundary markers: None explicitly defined to delimit user-provided values within the instruction template.\n
- Capability inventory: Uses
curlto perform network operations based on the processed input.\n - Sanitization: The workflow includes a specific step to 'Determine Minor Version' (e.g., extracting '1.34' from '1.34.1...'), which serves as an implicit validation and sanitization filter before the value is used in the URL.
Audit Metadata