GeoAI/ML Engineer
Pass
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: SAFENO_CODEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external geospatial imagery and datasets (such as Microsoft ML Buildings or Open Buildings). While this establishes a surface for indirect prompt injection via untrusted data, the instructions specifically mandate rigorous validation, visual spot-checking, and the use of per-class metrics to ensure model integrity.
- Ingestion points: External imagery files and third-party labeled datasets referenced in Phase 1.
- Boundary markers: None explicitly defined in the instructional text.
- Capability inventory: The skill lists tools for file processing (GDAL, Rasterio) and ML development (PyTorch, YOLO), but contains no executable code or scripts itself.
- Sanitization: Relies on human-in-the-loop visual validation and metrics-based performance monitoring.
- [NO_CODE]: The skill consists entirely of instructional markdown and persona definitions. It does not include executable scripts, shell commands, or automated installation routines.
Audit Metadata