gpu-document-processing
Pass
Audited by Gen Agent Trust Hub on Jul 4, 2026
Risk Level: SAFENO_CODE
Full Analysis
- [SAFE]: No malicious patterns, hardcoded secrets, or unauthorized behaviors were detected. The skill describes a secure architecture for offloading intensive tasks to isolated GPU environments.
- [EXTERNAL_DOWNLOADS]: Mentions well-known technology tools and services from NVIDIA, including NeMo Retriever, cuDF, and Triton. These are documented as architectural components for production deployments without including any automated download scripts.
- [PROMPT_INJECTION]: The skill defines a workflow for processing untrusted user data, which establishes an indirect prompt injection surface.
- Ingestion points: Large PDF files and document collections provided by users (SKILL.md).
- Boundary markers: None specified in the high-level architectural overview.
- Capability inventory: Text extraction, chunking, embedding generation, and tabular data parsing.
- Sanitization: No sanitization or validation methods are explicitly described for the ingested content.
Audit Metadata