office-to-md

Pass

Audited by Gen Agent Trust Hub on Jul 8, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill documentation includes instructions for installing the markitdown package via pip and executing it through both a command-line interface and Python subprocess-style interactions to convert documents.
  • [PROMPT_INJECTION]: The skill is designed to ingest and process data from external, potentially untrusted sources (Office documents and PDFs). This creates a vulnerability surface for indirect prompt injection, where malicious instructions hidden within a document could be processed by the agent after conversion to Markdown.
  • Ingestion points: The skill processes various Office formats (.docx, .xlsx, .pptx), PDFs, HTML, images, and ZIP archives as listed in the 'Supported Formats' table in SKILL.md.
  • Boundary markers: The provided code examples for conversion and archiving do not implement boundary markers or specific delimiters to isolate the converted content from the agent's core instructions.
  • Capability inventory: The skill utilizes file-system read and write capabilities (open(), pathlib.Path, markitdown.convert()) and integrates with an LLM client (anthropic) for vision-based tasks.
  • Sanitization: The example scripts perform direct conversion and storage of extracted text content without implementing sanitization, filtering, or validation of the text extracted from the source documents.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 8, 2026, 07:46 PM
Security Audit — agent-trust-hub — office-to-md