blockbench-new-model
Pass
Audited by Gen Agent Trust Hub on Sep 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill automates complex workflows by executing local command-line tools such as blender, ffmpeg, and npx. It generates and runs a render.cmd or render.sh script to coordinate parallel GPU rendering processes.\n- [EXTERNAL_DOWNLOADS]: The skill downloads the blockbench-mcp-plugin from the author's GitHub repository using npx. It additionally fetches modeling libraries like three and three-blockbench during the initial setup of the rendering engine.\n- [REMOTE_CODE_EXECUTION]: The skill setup involves running the blockbench-mcp-plugin directly from the author's GitHub repository using the npx -y github:jasonjgardner/blockbench-mcp-plugin command, which executes remote code locally.\n- [DYNAMIC_EXECUTION]: The skill dynamically generates Python scripts (import_and_fix.py, assemble.py, lights_cameras.py) to control Blender operations and creates batch/shell scripts for managing the render pipeline. These generated scripts are then executed as part of the normal project build process.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests user input to define 3D subjects, which are then processed by subagents, presenting a potential surface for indirect manipulation.\n
- Ingestion points: User intake form questions and the subject command argument in SKILL.md.\n
- Boundary markers: None identified to separate user-provided data from agent instructions.\n
- Capability inventory: Broad local capabilities including file system modification, command execution, and network access.\n
- Sanitization: No specific input sanitization or validation logic is documented for user-supplied subject descriptions.
Audit Metadata