blender
Pass
Audited by Gen Agent Trust Hub on Sep 5, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill uses instructional markers such as "The single rule that outranks everything" to enforce specific operational priorities, such as visual verification of renders over numerical data.
- [DYNAMIC_EXECUTION]: The skill provides Python code snippets designed for execution within the Blender environment via the
execute_blender_codetool. It correctly identifies that the environment's sandbox is intended as guidance rather than strict containment and provides defensive coding patterns to prevent data corruption or accidental file modification. - [COMMAND_EXECUTION]: The skill instructs the agent on how to invoke Blender's headless mode via the shell (e.g.,
blender -b), including handling arguments and background processes. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external Blender scenes and assets, which represents a vulnerability surface for indirect instructions.
- Ingestion points: Reading and rendering
.blendfiles and external assets. - Boundary markers: None explicitly defined for the generation of Blender scripts.
- Capability inventory: Includes arbitrary Python execution within Blender (
bpyAPI) and shell command execution for headless renders. - Sanitization: Recommends visual verification of output images (renders) to detect anomalies that automated measurements might miss.
- [EXTERNAL_DOWNLOADS]: The skill references the official Blender Foundation repositories and documentation (projects.blender.org) as sources for the MCP server and related tools.
Audit Metadata