product-ad-shot-design
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data in the form of third-party prompt excerpts from
references/cases.md. This creates a theoretical surface for indirect prompt injection where instructions hidden within these examples could attempt to influence the agent's output. \n - Ingestion points:
references/cases.md(containsPrompt excerptfields from 8+ different cases).\n - Boundary markers: The skill implements a 'reference contract' (Preserve/Replace/Avoid) and a 'Required evidence gate' to structure how the data is used.\n
- Capability inventory: The agent primarily generates text-based video prompts and shot briefs; it does not have autonomous execution capabilities that would escalate the risk of this surface.\n
- Sanitization: Explicit instructions are provided to 'distinguish observed evidence from your own recommendation' and to avoid impersonating creators.\n- [EXTERNAL_DOWNLOADS]: The skill references multiple external media assets (MP4, JPG) and case pages hosted on
goodcase.aiandx.com. These are used as visual references for the prompt design workflow. The agent is instructed to inspect these using its internal tools (browser, image/video viewer). These are legitimate resources for the skill's stated purpose and do not involve executable code.
Audit Metadata