ocr-and-documents

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill parses and outputs content from untrusted external documents (PDF, DOCX, PPTX), creating a potential vector for indirect prompt injection where malicious instructions embedded in a document could influence the agent's behavior.
  • Ingestion points: Files are read and processed in scripts/extract_marker.py and scripts/extract_pymupdf.py.
  • Boundary markers: The script scripts/extract_pymupdf.py includes basic page delimiters (--- Page X/Y ---) which offer limited separation between document content and system instructions.
  • Capability inventory: The skill can write files (images) to the local disk and returns large volumes of extracted text to the agent context.
  • Sanitization: The extracted text is not sanitized or filtered for potential instructional content before being processed by the agent.
  • [EXTERNAL_DOWNLOADS]: The marker-pdf dependency fetches approximately 2.5GB of model weights from Hugging Face during its initial setup. This is a documented behavior of the underlying library for performing high-quality OCR.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 12:47 PM
Security Audit — agent-trust-hub — ocr-and-documents