The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill provides tools for running arbitrary Python scripts and entry points within the Blender environment.\n
Evidence: scripts/run_script.py and scripts/run_entrypoint.py enable the execution of file-based scripts and specific module functions.\n- [REMOTE_CODE_EXECUTION]: The install_addon tool allows the installation of new Python code into the application.\n
Evidence: scripts/install_addon.py installs and optionally enables add-ons from .py or .zip files, which persists new executable content.\n- [COMMAND_EXECUTION]: The attach_project tool allows modification of the Python environment's search paths.\n
Evidence: scripts/attach_project.py appends user-provided directory paths to sys.path, enabling the loading of modules from outside the standard locations.\n- [INDIRECT_PROMPT_INJECTION]: The skill exposes a vulnerability surface where data ingested from the UI or files can interact with high-privilege execution tools.\n
Ingestion points: UI metadata is ingested via capture_ui_snapshot.py and find_ui_elements.py.\n
Boundary markers: The skill does not implement boundary markers or instructions to ignore embedded commands in the processed UI metadata.\n
Capability inventory: The skill includes powerful capabilities such as arbitrary script execution (run_script) and add-on management (install_addon, remove_addon).\n
Sanitization: No sanitization is performed on the UI data or file paths before they are used in operations that could affect the system state.\n- [DATA_EXFILTRATION]: The start_debug_server tool opens a network listener on the local machine.\n
Evidence: scripts/start_debug_server.py starts a debugpy server on a specified host and port (defaulting to 127.0.0.1:5678), which could allow remote access to the running process if exposed or misconfigured.

blender-dev