editorial-portrait-composition-by-1enbgh4
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill is designed to fetch and inspect media and textual descriptions from external websites to use as reference material for prompt generation.
- Evidence: The file
references/cases.mdcontains multiple links togoodcase.ai,media.goodcase.ai, andx.com(e.g.,https://goodcase.ai/cases/nano-banana-created-with-google-gemini-nano-banana-pro-3209562796f4). - Evidence: The
WorkflowinSKILL.mdinstructs the agent to "Inspect its finished media, summary, and prompt excerpt" using available browser or image tools. - [INDIRECT_PROMPT_INJECTION]: The skill processes data from third-party websites, creating a surface for potential indirect prompt injection if the external content contains malicious instructions.
- Ingestion points: The agent is instructed to read
references/cases.mdand then browse/inspect the external URLs listed within that file (inSKILL.mdunder "Required evidence gate"). - Boundary markers: The skill uses a "reference contract" (Preserve/Replace/Avoid) to structure the ingestion and provides instructions to "Distinguish observed evidence from your own recommendation."
- Capability inventory: The agent possesses capabilities to browse the web, inspect images/media, and generate text-based image prompts.
- Sanitization: There is no explicit sanitization or filtering logic mentioned for the content retrieved from external URLs before it is incorporated into the prompt generation process.
Audit Metadata