producto
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided photos and external URLs, which represents a potential surface for indirect prompt injection if the source media or metadata contains hidden instructions.\n
- Ingestion points: Untrusted data enters the agent context via the media.subir tool, which accepts both local file uploads and remote URLs, as well as configuration from project files like marca.json.\n
- Boundary markers: The skill instructs the agent to use structured prompt blocks and a fidelity anchor defined in the direccion-de-arte skill to maintain focus on the real product and minimize instruction override.\n
- Capability inventory: The skill utilizes tools for image editing, background removal, image expansion, scaling, and cost consultation.\n
- Sanitization: A strict manual review process is required, where the agent must compare the generated image side-by-side with the original photo to detect and discard results with altered labels, logos, or product geometry.
Audit Metadata