neqsim-document-intelligence-extraction
Pass
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from various document formats (PDF, DOCX, XLSX, images, and P&IDs), which presents a potential surface for indirect prompt injection.
- Ingestion points: The workflow ingestion process involves recursive file inventory and extraction from paths provided in
source_path. - Boundary markers: The skill uses a structured
EvidenceContractto separate original source text from normalized values and extraction methods, preventing the agent from conflating data with instructions. - Capability inventory: The skill utilizes the
document_intelligence_extractionlibrary to perform file parsing, coordinate-aware OCR, and multimodal vision inference. - Sanitization: The workflow enforces human-review gates for any safety-critical data or findings with a confidence score below 0.85, mitigating the risk of the agent acting on malicious instructions hidden within document content.
- [EXTERNAL_DOWNLOADS]: The documentation and code samples reference the official Equinor GitHub repository for the NeqSim project. These links provide access to relevant technical guides and source code maintained by the skill's author.
Audit Metadata