mobility-gen
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes local script execution via
run_script()and the Isaac Simpython.shshell launcher to coordinate physics simulations and sensor rendering tasks.- [INDIRECT_PROMPT_INJECTION]: The skill ingests external simulation parameters and environment data from files such asmap.yaml,config.json, and USD stages. This constitutes an attack surface where maliciously crafted simulation metadata could influence the agent's behavior. - Ingestion points: Occupancy map YAML files (
omap_yaml) and recording configuration JSON files processed inscripts/record_trajectories.pyandscripts/replay_custom_robot.py. - Boundary markers: Absent; the skill relies on standard format parsing without specific delimiters for external content.
- Capability inventory: Extensive simulation control including
SimulationApplifecycle management, file system access for recording datasets (os.makedirs,np.savez), and rendering operations. - Sanitization: The skill performs standard YAML and JSON loading without specific validation or filtering against embedded natural language instructions.
Audit Metadata