The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests metadata and geometry data from the active Blender scene (such as object and layer names) to generate reports and manage drawings. This represents a potential surface for indirect prompt injection if the Blender project file originates from an untrusted source.
Ingestion points: bx_gp.py reads object, layer, and material metadata from bpy.context.scene in functions like report() and drawing_stats().
Boundary markers: No explicit boundary markers or 'ignore' instructions are used when processing scene data.
Capability inventory: The skill can write files to the local system (render_frame, render_strip, onion) and execute Blender operators (edit_op, gui_trim).
Sanitization: Scene-derived strings are not explicitly sanitized before being used in output reports or as identifiers.
[DYNAMIC_EXECUTION]: The skill dynamically modifies the Python path to import peer scripts and uses dynamic attribute access to invoke Blender operators.
Evidence: bx_gp.py appends a sibling directory (scenario-blender-expert/scripts) to sys.path to load dependencies from the same author.
Evidence: The edit_op function in bx_gp.py uses getattr to dynamically call operators within the bpy.ops namespace based on string arguments.

scenario-blender-grease-pencil