robot-perception
Pass
Audited by Gen Agent Trust Hub on Sep 16, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is entirely educational and functional, providing code implementations for standard robotic perception tasks.
- [INDIRECT_PROMPT_INJECTION]: The skill defines pipelines for processing external sensor data (images, depth maps, and point clouds).
- Ingestion points: Data is ingested through sensor interfaces such as
camera.capture()androbot.get_ee_pose()inSKILL.md. - Boundary markers: As the skill handles raw binary sensor data (uint8/float arrays), text-based boundary markers are not applicable.
- Capability inventory: The skill utilizes image processing (OpenCV), point cloud registration (Open3D), and local file I/O for calibration parameters using
cv2.FileStorage. - Sanitization: The provided code includes validation logic, such as filtering invalid depth values, checking reprojection errors, and removing statistical outliers, which mitigates risks associated with processing malformed sensor data.
- [COMMAND_EXECUTION]: All code examples use standard Python libraries and well-known robotics/vision packages (OpenCV, Open3D) for their intended purposes without performing arbitrary shell commands or privilege escalation.
Audit Metadata