document-conversion

Pass

Audited by Gen Agent Trust Hub on May 20, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The workflow directs the agent to utilize standard command-line utilities and libraries such as pdftotext, PyMuPDF, Pandoc, and Docling for document parsing. These tools are used for their primary intended purpose within the local environment to facilitate document conversion.
  • [EXTERNAL_DOWNLOADS]: The documentation references the use of hosted converters (e.g., MinerU) and explicitly advises the user to record external dependencies and consider security/API risks associated with third-party services, demonstrating a security-aware policy.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (PDFs, DOCX, Scanned documents) which is a known vector for indirect prompt injection.
  • Ingestion points: Raw documents are ingested from sources/pdfs/, reports/, and data/raw/ (referenced in SKILL.md and references/document-conversion-policy.md).
  • Boundary markers: The skill lacks explicit instructions to the agent to ignore instructions embedded within the source documents during the conversion process, though it mandates manual quality checks.
  • Capability inventory: The agent has capabilities to write to the local filesystem (sources/, wiki/, docs/) and execute system commands for conversion (documented in references/repository-contract.md).
  • Sanitization: There is no automated sanitization or filtering of document content; however, the 'Quality Gate' in SKILL.md requires manual verification of extracted evidence against the native source file, which acts as a human-in-the-loop mitigation.
Audit Metadata
Risk Level
SAFE
Analyzed
May 20, 2026, 09:06 AM