ai-ad-creative
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides numerous examples using
curlto interact with the official each::labs API athttps://eachsense-agent.core.eachlabs.run. These commands are used for legitimate data transmission and do not involve piping remote content to a shell. - [INDIRECT_PROMPT_INJECTION]: The skill supports processing external images via the
image_urlparameter. This introduces a surface where malicious instructions could be embedded in image metadata or visual content to influence the AI agent's behavior. - Ingestion points:
image_urlfield in the multimodal message payload (SKILL.md). - Boundary markers: None implemented in the provided examples; content is passed directly to the API.
- Capability inventory: Network access via
curlandWebFetch(SKILL.md). - Sanitization: No evidence of input sanitization or image validation in the provided instructions.
Audit Metadata