linkfox-multimodal-extract-attributes
Pass
Audited by Gen Agent Trust Hub on Apr 22, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill processes untrusted product data through a multimodal AI, which is a vector for indirect prompt injection.
- Ingestion points: Untrusted product lists, including titles and image URLs, are ingested via the 'refResultData' parameter in SKILL.md.
- Boundary markers: There are no delimiters or explicit instructions to the AI to ignore potential instructions embedded within the processed image content or metadata.
- Capability inventory: The skill performs network requests to 'tool-gateway.linkfox.com' using the 'scripts/multimodal_extract_attributes.py' script.
- Sanitization: No sanitization or validation is applied to the incoming product data before it is transmitted to the multimodal model.
Audit Metadata