blender-previs-storyboard
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted natural language data (story beats and shot intents) which are then used to generate shot lists, stamps for renders, and audit reports. Malicious instructions embedded in this text could potentially influence the agent's behavior during review or auditing tasks.
- Ingestion points: The BEATS list and SHOTS configuration data ingested by P.shot_list in scripts/bx_previs.py and SKILL.md.
- Capability inventory: The skill can modify Blender project data and write various files (.png, .mp4, .json) to the local file system using the Blender API.
- Boundary markers: The skill does not implement specific delimiters or instructions to ignore embedded commands within the ingested story text.
- Sanitization: Input strings are used without validation or escaping before being rendered into images or included in audit logs.
- [EXTERNAL_DOWNLOADS]: The skill incorporates external Python dependencies and relies on sibling modules from the same vendor.
- The script scripts/bx_previs.py imports the numpy package, which is a well-known scientific library.
- The skill configuration and scripts reference a sibling module bx_review and expected scripts from a blender-expert skill directory.
Audit Metadata