ocr-and-documents

Pass

Audited by Gen Agent Trust Hub on Jun 16, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches machine learning models (~2.5GB) from Hugging Face's official repositories (huggingface.co) during the initial execution of marker-pdf. This is a standard operation for this library and targets a well-known service.\n- [PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection.\n
  • Ingestion points: Extracts text and data from user-provided files (PDF, DOCX, PPTX, XLSX, HTML, EPUB) and images via scripts/extract_marker.py and scripts/extract_pymupdf.py.\n
  • Boundary markers: No delimiters or specific safety instructions are used to separate the extracted document content from the agent's system instructions.\n
  • Capability inventory: The skill employs web_extract, web_search, and execute_code tools, which could be exploited if the agent follows malicious instructions embedded within processed documents.\n
  • Sanitization: No automated filtering or sanitization is performed on the extracted text before it is returned to the agent context.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 16, 2026, 12:58 AM
Security Audit — agent-trust-hub — ocr-and-documents