game-image-generation-rules

Pass

Audited by Gen Agent Trust Hub on Sep 8, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes image assets (PNG, SVG) and uses vision models to evaluate them, creating a surface for indirect prompt injection.
  • Ingestion points: Assets are processed by scripts like scripts/inspect_asset.py and reviewed in the vision evaluation loop.
  • Boundary markers: There are no explicit instructions for the AI to ignore potential instructions embedded within the images or their metadata.
  • Capability inventory: The workflow includes image generation, editing, and vision-based decision making.
  • Sanitization: While the SVG script flags scripts and external references, it does not provide comprehensive sanitization of assets before they are presented to the vision model.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 8, 2026, 11:35 AM
Security Audit — agent-trust-hub — game-image-generation-rules