skills/datadrivenconstruction/ddc_skills_for_ai_agents_in_construction/image-to-data/Gen Agent Trust Hub
image-to-data
Pass
Audited by Gen Agent Trust Hub on Mar 5, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill has a standard architectural surface for indirect prompt injection common to all vision-based agents. Because it extracts text from user-provided images via OCR and processes it, there is a theoretical risk that instructions embedded within an image could influence agent behavior.
- Ingestion points: Image files (JPG, PNG, TIFF, BMP) read from the filesystem.
- Boundary markers: No specific delimiters are used to wrap OCR-extracted text within the internal processing logic.
- Capability inventory: The skill has filesystem read access and network access to communicate with AI vision providers (OpenAI/Claude).
- Sanitization: No explicit sanitization or filtering of OCR output is defined in the instructions.
Audit Metadata