scenario-orbit-views
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions direct the agent to execute local Python scripts (
compose_layout.py,match_view.py,studio_shadow.py) and perform Blender automation via the command line (blender -b ... --python render_grounded.py). This is the core functionality for processing 3D meshes and images. - [INDIRECT_PROMPT_INJECTION]: Step 6 of the
SKILL.mduses a prompt template that interpolates untrusted user inputs (e.g.,{details},{finish}) directly into instructions sent to an image generation model. - Ingestion points: Untrusted text data enters the agent context via variables in the
SKILL.mdprompt template. - Boundary markers: Absent; user-controlled variables are not delimited or wrapped in safety instructions.
- Capability inventory: The skill possesses file writing, local script execution, and network download capabilities.
- Sanitization: There is no evidence of escaping or validation for the text data before interpolation.
- [EXTERNAL_DOWNLOADS]: The skill uses
curl -Lto download 3D GLB assets from URLs generated during the MCP workflow. These downloads originate from the vendor's platform context as part of the asset creation pipeline.
Audit Metadata