The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is vulnerable to Python code injection because it constructs code strings using unvalidated input.
Evidence: In scripts/polyhaven-material-swap.js, variables such as objectName, samples, renderWidth, and renderHeight are interpolated directly into Python code templates.
Vulnerability: The script fails to sanitize or escape these variables before they are embedded in the Python string. For example, the code obj = bpy.data.objects.get("${objectName}") allows an attacker to terminate the string and inject arbitrary Python commands using a payload like Mesh"); import os; os.system("..."); #.
Risk: Maliciously crafted command-line arguments can lead to arbitrary code execution within the context of the Blender application.
[REMOTE_CODE_EXECUTION]: The skill transmits dynamically generated code over a network socket to a listening service.
Evidence: The script uses the Node.js net module to connect to localhost:9876 and send JSON payloads containing Python code in the execute_code field.
Capability: The execute_code functionality on the Blender MCP server is highly permissive, allowing any valid Python code to run with the permissions of the Blender process.
[DATA_EXFILTRATION]: The code injection vulnerability creates a significant surface for data exfiltration.
Attack Surface: An attacker could use the injection vector to read sensitive local files or environment variables and exfiltrate them using Python's networking libraries or the existing socket connection.

polyhaven-material-swap