The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/run_in_blender.py uses subprocess.run to invoke the Blender binary for headless operations.\n
Evidence: Line 85 in scripts/run_in_blender.py executes a command constructed from environment paths and user-provided script arguments.\n- [DYNAMIC_EXECUTION]: The skill facilitates the execution of arbitrary Python code within the Blender environment through several mechanisms.\n
Evidence: scripts/run_in_blender.py includes a --expr flag that passes user-defined strings directly to Blender's --python-expr parameter.\n
Evidence: SKILL.md instructs the agent to use importlib.reload to dynamically update and execute scripts during live sessions.\n- [INDIRECT_PROMPT_INJECTION]: The skill exposes a vulnerability surface where untrusted data could influence agent actions or system state by injecting malicious instructions into processed Blender code.\n
Ingestion points: The run_in_blender.py script and the live MCP execute_blender_code tool ingest strings that are executed as Python code within the Blender process.\n
Boundary markers: The instructions do not define clear boundaries or 'ignore' directives for data processed by these tools to distinguish between intended code and embedded instructions.\n
Capability inventory: The skill can execute shell commands via subprocess and has full access to the local machine's filesystem and networking capabilities via the standard Python and Blender bpy APIs.\n
Sanitization: There is no evidence of input sanitization or sandboxing logic to prevent malicious code from being injected into the Blender execution context.

blender-core