docx

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

• Third-party content exposure detected (high risk: 0.75). In the “Redlining workflow” path, the skill ingests outsider-authored free text from the user-supplied DOCX by converting it to markdown with pandoc --track-changes=all ... -o current.md, which then becomes LLM-readable context for change identification.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill explicitly instructs installing system packages with commands like "sudo apt-get install ..." and global installs (npm -g), which directs the agent to obtain elevated privileges and modify the host system state.