The Agent Skills Directory

[SAFE]: The skill employs defensive programming by using the defusedxml library for XML parsing across its components, which prevents XML External Entity (XXE) attacks. The extraction process in ooxml/scripts/unpack.py includes validation logic to protect against path traversal (ZipSlip).
[COMMAND_EXECUTION]: It invokes local system binaries soffice and git for document validation and text diffing respectively. These calls are executed with fixed argument lists and controlled paths, minimizing injection risk.
[EXTERNAL_DOWNLOADS]: All external dependencies, including pandoc and the docx library, are standard, well-known tools used according to their intended primary purpose.
[PROMPT_INJECTION]: The skill processes untrusted Word documents which could contain indirect prompt injections. Data enters the agent context through pandoc extraction or raw XML reading in scripts/document.py. While boundary markers for this content are minimal, the skill provides a capability inventory limited to its document editing library and local tool calls, with sanitization performed via secure XML parsers.

docx-official