wireframe-to-3d
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a local Python script
scripts/wireframe_analyzer.pyvia the bash tool to process user-supplied images. Additionally, it uses themcp__blender__execute_blender_codetool to run dynamically generated Python code within a Blender instance to build 3D models. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted binary data in the form of PNG images provided by the user.
- Ingestion points: User-provided PNG files are processed by the
scripts/wireframe_analyzer.pyscript as defined inSKILL.md. - Boundary markers: The skill converts unstructured image data into structured JSON containing Bezier control points. This structural transformation acts as a boundary, but the skill lacks explicit instructions to the agent to treat the resulting geometric data as untrusted when generating Blender Python code.
- Capability inventory: The skill can execute shell commands, perform local file system operations (reading files and writing to
/tmp), and execute arbitrary Python code in the Blender environment through the provided MCP tool. - Sanitization: The extraction of contours and Bezier points provides a functional filter, but there is no explicit sanitization to prevent the agent from interpreting adversarial patterns within the wireframes that could be designed to influence the code generation phase.
Audit Metadata