The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates and executes Python scripts within Blender and C# scripts within Unity to perform model export and scaling operations.
In Step 2, Python code is generated to export files via bpy.ops.export_scene.fbx or bpy.ops.export_scene.gltf.
In Step 4, C# code is generated to find game objects and modify their localScale based on bounds calculation.
[COMMAND_EXECUTION]: The skill uses execute_code and mcp__blender__execute_blender_code to run arbitrary instructions in the host applications. This capability allows for complex operations that could be misused if the generated strings are compromised.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect injection because it ingests external data (Blender object names and Unity project paths) and interpolates them directly into executable code templates.
Ingestion points: Data enters through mcp__blender__get_object_info (object names) and mcpforunity://editor/state (project paths).
Boundary markers: None identified. The instructions do not warn the agent to sanitize these inputs before interpolation.
Capability inventory: The skill uses mcp__blender__execute_blender_code (Python execution in Blender) and execute_code (C# execution in Unity).
Sanitization: There is no evidence of escaping or validating the <asset name> or <temp path> variables before they are placed into code strings (e.g., GameObject.Find("<asset name>")). If an object name contained characters like "); followed by malicious C# code, it could trigger unexpected behavior in the Unity Editor.
[DATA_EXFILTRATION]: The skill explicitly notes it does not send data over the bridge, which is a positive security practice. However, it still interacts with the local filesystem and temporary directories, which could be monitored or used as a pivot point for exfiltration if the execution environment is compromised.

blender-to-unity