neural-reality-capture
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill provides comprehensive technical documentation and procedural guidance for 3D reconstruction and neural radiance fields. The content is educational and professional in nature.
- [SAFE]: External references target official documentation and project pages for reputable academic research (Inria, UC Berkeley) or established industry standards and tools (COLMAP, Khronos Group, Agisoft, RealityCapture, and OpenUSD). These are well-known technical resources.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external media (images, video) and associated metadata (EXIF). While this constitutes an ingestion surface for untrusted data, the skill does not provide or invoke dangerous capabilities such as arbitrary code execution or network exfiltration that would make this surface exploitable.
- Ingestion points: User-provided photographs, video observations, and EXIF metadata (SKILL.md).
- Boundary markers: The skill utilizes 'Evidence Labels' (e.g., 'Paper fact', 'Standard fact') to categorize information sources, though it lacks explicit delimiters for raw external data.
- Capability inventory: No shell commands, file-write operations, or network exfiltration scripts are defined within the skill files.
- Sanitization: No specific input sanitization or filtering logic is provided for external data strings.
Audit Metadata