style-loop
Pass
Audited by Gen Agent Trust Hub on Oct 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill follows a highly structured and documented workflow for art direction. It manages state locally within a dedicated
.style-loop/directory and focuses on project asset management and style consistency. - [INDIRECT_PROMPT_INJECTION]: The skill processes external visual data, such as game assets and screenshots, which constitutes an attack surface for indirect prompt injection.
- Ingestion points: Reads existing game assets, screens, and project reference files from the local directory into the agent context (e.g.,
references/workflow.md). - Boundary markers: Documentation in
references/judge.mdspecifies the use of 'neutral labels' and 'blind packets' to isolate evaluation content, which provides functional boundaries for data processing. - Capability inventory: The skill utilizes image generation tools and performs file system operations to write structured Style Kits and Bibles.
- Sanitization: Documentation explicitly instructs the agent to treat text embedded in art or reference images as artifact content rather than instructions (per
references/judge.md). - [EXTERNAL_DOWNLOADS]: The documentation provides standard installation instructions using
git clonefrom the author's official GitHub repository and thenpx skillscommand from the well-known Vercel Labs repository. These references are legitimate and originate from the vendor's own infrastructure or established services.
Audit Metadata