The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from the local filesystem and asset libraries, which introduces a potential surface for indirect prompt injection if an attacker controls file names or metadata.
Ingestion points: The search_assets tool (via scripts/search_assets.py) reads file paths, names, and metadata from the filesystem and Blender's internal library registrations.
Boundary markers: The skill does not define explicit delimiters or instructions for the agent to ignore potentially malicious content embedded in file metadata or names.
Capability inventory: The skill uses Bash and Read for search operations but is marked as read_only and destructive: false in its tool definition, limiting the impact of any potential injection.
Sanitization: No explicit sanitization logic is provided to filter or escape filenames or asset metadata before they are returned to the agent context.

blender-asset-source