robolab-scenegen
Pass
Audited by Gen Agent Trust Hub on Jun 21, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local Python scripts (
settle_scenes.py,generate_catalog.py) and shell commands to perform physics simulation and manage object metadata. It includes automated discovery of the simulation environment's Python interpreter (e.g., within Conda environments) and executes a dynamically generated Python script to solve spatial placement constraints. - [PROMPT_INJECTION]: The skill processes untrusted user input to define scene layouts, which presents an indirect prompt injection surface.
- Ingestion points: User-provided natural language 'Scene description' and 'custom object directory' paths (SKILL.md).
- Boundary markers: No explicit delimiters are used for the description text; the input is interpolated into a JSON structure for the solver script.
- Capability inventory: File system write access (USDA scene generation), Bash execution (Python solver, settlement scripts, catalog utilities).
- Sanitization: The skill validates that output filenames end with the '.usda' extension and includes checks to prevent overwriting existing files.
Audit Metadata