docx

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.80). The skill explicitly ingests and processes arbitrary user/third‑party .docx files (e.g., via pandoc conversion "pandoc --track-changes=all", the unpack.py script which extracts and pretty‑prints office_file, and the Document class which reads word/document.xml), so the agent will read and interpret untrusted document content that could carry indirect prompt injection.