mira-ugc
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input, such as product descriptions and labels, and incorporates it into prompts for generative models and captions for social media. This creates a potential surface where adversarial instructions in the input could influence the output media or the agent's behavior during publishing steps.\n
- Ingestion points: User-provided product details and label text used in the Doctrine and Interview sections (SKILL.md).\n
- Boundary markers: The skill contains instructions to the agent to keep product labels exactly as given, which serves as a prompt instruction but does not provide a robust security boundary.\n
- Capability inventory: The skill utilizes media generation tools (generate_video, generate_image) and mentions social media publishing capabilities for platforms like Instagram and Threads.\n
- Sanitization: No evidence of explicit sanitization or validation of user-provided content is present.\n- [SAFE]: The use of Cyrillic text for keywords like 'отзыв' (review) and 'распаковка' (unboxing) is contextually appropriate for supporting multilingual user requests and does not involve obfuscation or homoglyph attacks.\n- [SAFE]: References to internal project tools and model-specific configurations for generating media and marketing content are consistent with the skill's stated purpose and do not represent security risks.
Audit Metadata