The Agent Skills Directory

[SAFE]: The skill defines a controlled environment for running Blender background jobs, utilizing snapshot isolation to ensure that background tasks are bound to specific scene states. It explicitly restricts file writes to approved job directories.
[INDIRECT_PROMPT_INJECTION]: The skill involves processing external data, such as Blender scenes and local assets. While this is an ingestion point for untrusted data, the skill includes several security controls:
- Ingestion points: External Blender scenes and assets referenced in the request example.
- Boundary markers: The SKILL.md includes a 'Rules 与能力边界' (Rules and Boundaries) section and a dedicated acceptance-checklist.md to guide the agent.
- Capability inventory: Execution is limited to a child Blender process with restricted write access.
- Sanitization: The workflow requires verification of frame hashes, scene revisions, and terminal receipts to ensure data integrity.

blender-background-jobs