The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill accepts external input (object names and simulation parameters) to configure the Blender scene, creating a potential surface for indirect prompt injection.
Ingestion points: The scripts/sim_setup.py script takes object names (--active, --passive, --cloth, --colliders) and key-value pairs (--set) via command-line arguments.
Boundary markers: Absent; the skill does not implement delimiters or safety instructions to prevent the agent from misinterpreting instructions embedded within these data strings.
Capability inventory: The skill has the ability to modify the Blender scene, configure modifiers, and save the main file (bpy.ops.wm.save_as_mainfile) as evidenced in scripts/sim_setup.py and scripts/_bpyx.py.
Sanitization: Basic type conversion for values is performed in the _num helper, and object names are validated against bpy.data.objects, but property names and attributes are handled dynamically.
[METADATA_POISONING]: There is a discrepancy between the author name listed in the skill metadata and the provided author context.
Evidence: The YAML frontmatter in SKILL.md lists the author as blender-agent-skills, which does not match the provided author context el4cteo. This constitutes misleading metadata, although no malicious instructions are present in the metadata fields.

blender-simulation