qwen-mm-plugins-blender
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill uses
importlib.import_modulewith computed paths to interface with Blender add-ons. While this provides flexibility across different installation types (legacy vs. extensions), it involves loading code from namespaces determined at runtime. - [EXTERNAL_DOWNLOADS]: Upon the first invocation of a tool, the skill is designed to automatically download and install a pinned version of the Blender binary (~300MB) for Linux-x86_64 systems. It also allows for the retrieval of assets and materials from external repositories like Sketchfab and Polyhaven.
- [COMMAND_EXECUTION]: The core functionality of the skill relies on the
execute_blender_codetool, which allows the agent to execute arbitrary Python scripts within a running Blender instance to manipulate the 3D scene. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted metadata from external asset libraries (Sketchfab, Polyhaven). This data enters the agent's context and could influence the generation of Blender Python scripts, representing an indirect injection surface.
- Ingestion points: Search results and asset metadata from the Sketchfab and Polyhaven tools.
- Boundary markers: The instructions lack explicit guidance on using delimiters to separate external content.
- Capability inventory: High capability via the Blender Python API and local file system access for imports.
- Sanitization: No sanitization of third-party asset metadata is mentioned.
Audit Metadata