The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses curl to send Python code to a local server, python3 to execute a management script (start_server.py), and pkill for process management.
[DYNAMIC_EXECUTION]: Instructs the agent to use exec(open("/tmp/blender_script.py").read()) to execute Python scripts within the Blender environment. This is the primary mechanism for the skill's operation.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from the Blender environment (scene objects, properties) and logs, which are then used to generate subsequent Python commands, creating a surface for indirect instruction injection if external or untrusted data is present in the 3D scene.
Ingestion points: Reading Blender scene data (bpy.data.objects.keys()) and reading log files (output/agent.log, ~/Library/Logs/DiagnosticReports/Blender-*.ips).
Boundary markers: None explicitly defined in the instruction set.
Capability inventory: Full Python execution within the Blender process (exec), shell command execution (curl, pkill), and file system writes (cat > /tmp/...).
Sanitization: No explicit sanitization or validation of data retrieved from the scene before it is interpolated into Python scripts for execution.

blender