blender-creation
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFECOMMAND_EXECUTIONPRIVILEGE_ESCALATIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill relies on shell command execution for its primary functions.
scripts/asset_build_utils.pyusessubprocess.runto invoke the Blender binary for rendering and diagnostics.SKILL.mdinstructs the agent to useuv runto execute various processing and verification scripts. - [PRIVILEGE_ESCALATION]: The instructions in
SKILL.mdinclude a routine to check for the Blender binary and executesudo apt-get update && sudo apt-get install -y blenderif it is missing, requiring administrative privileges for software installation. - [EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch external resources.
scripts/asset_build_utils.pyusesurllib.request.urlopento download PBR texture maps fromambientcg.com(a well-known asset repository). Additionally,SKILL.mddirects the agent to download reference images from user-provided URLs usingcurl. - [DYNAMIC_EXECUTION]: The modeling pipeline involves generating Python scripts at runtime (e.g.,
build_asset.py,verify_and_render.py) which are then executed within the Blender environment or viauv. This is a standard mechanism for programmatic 3D content creation but constitutes dynamic code generation. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from product URLs using a
browsersubagent to extract descriptive metadata. This represents an attack surface where malicious instructions embedded in external web content could potentially influence agent behavior during the modeling process. - Ingestion points: External product pages accessed via the
browsersubagent as specified inSKILL.md(§1.1). - Boundary markers: Not explicitly defined in the instructions for webpage content processing.
- Capability inventory: File system access, shell command execution (
subprocess.run), and network requests (urllib.request.urlopen). - Sanitization: No specific sanitization or escaping logic is described for the extracted metadata prior to its use in generating modeling parameters.
Audit Metadata