The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/office/soffice.py dynamically generates and compiles a C shared library (lo_socket_shim.so) at runtime using gcc. This library is then loaded into the soffice (LibreOffice) process via the LD_PRELOAD environment variable to intercept and modify standard system calls related to Unix domain sockets.
[COMMAND_EXECUTION]: The skill executes several powerful external command-line utilities to perform its tasks, including gcc (for shim compilation), soffice (for document conversion and change management), git (for document diffing), pandoc (for text extraction), and pdftoppm (for image generation).
[PROMPT_INJECTION]: The skill inherently processes untrusted data by reading and manipulating user-provided Word documents (.docx, .doc). Content extracted via pandoc enters the agent's context, creating a surface for indirect prompt injection where instructions hidden within a document could attempt to influence agent behavior. While the skill correctly uses defusedxml for most XML parsing, the logic in scripts/office/validators/redlining.py and scripts/office/helpers/simplify_redlines.py relies on the standard xml.etree.ElementTree library, which does not provide the same level of protection against XML External Entity (XXE) vulnerabilities when processing untrusted input.

docx