lookdev-auto
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for ingesting model-generated ratings and suggestions to drive an automated rendering loop, creating a surface for indirect prompt injection.
- Ingestion points: Visual artifacts and model outputs (JSON ratings/suggestions) described in the rendering loop in
SKILL.md. - Boundary markers: The instructions recommend providing an explicit rubric and requesting structured JSON output to constrain responses.
- Capability inventory: The skill is documentation-only and contains no scripts; however, the workflow involves triggering rendering tools and model inference.
- Sanitization: Instructions suggest using regex to extract JSON from model responses, but do not specify input validation for the suggested parameter values.
Audit Metadata