Product Video Ads
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a surface for indirect prompt injection as it processes untrusted user-supplied data such as product briefs, brand guides, and audience descriptions.
- Ingestion points: The 'Required inputs' section in SKILL.md specifies that the agent should ingest external product/SKU identity, brand guides, and asset rights information.
- Boundary markers: The skill mitigates risks by instructing the agent to create a 'product preservation block' in Phase 1 to isolate and lock product invariants.
- Capability inventory: The skill utilizes muapi tools for media generation (
media.generate_video,media.generate_image) and file operations (media.upload_file) as listed in the frontmatter. - Sanitization: The instructions include a manual sanitization step, requiring the agent to mark unverified certifications or prices as 'unapproved' and request a source rather than generating plausible text.
- [SAFE]: The skill does not contain any evidence of prompt injection, credential exfiltration, persistence mechanisms, or unauthorized command execution. All media generation capabilities are scoped within the provided platform connections.
Audit Metadata