The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/office/soffice.py contains a hardcoded C source string that is written to a temporary file and compiled at runtime using gcc. The resulting shared library (lo_socket_shim.so) is used as a technical shim for LibreOffice.
[DYNAMIC_EXECUTION]: The script scripts/accept_changes.py generates a StarBasic macro (ACCEPT_CHANGES_MACRO) and writes it to a temporary LibreOffice user profile. This macro is then executed via the soffice command to programmatically accept all tracked changes in a document.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill uses LD_PRELOAD to inject the dynamically compiled C shim into the soffice process. This technique shims low-level socket functions (socket, listen, accept) to bypass environmental restrictions on AF_UNIX sockets.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted Word documents provided by users. This creates an attack surface where malicious instructions could be embedded in the document XML.
Ingestion points: scripts/office/unpack.py and scripts/office/validate.py (via zip extraction and XML parsing).
Boundary markers: XML structure and pretty-printing; however, the agent is instructed to edit raw XML directly.
Capability inventory: Extensive use of subprocess.run to call soffice, gcc, git, pandoc, and pdftoppm across multiple script files.
Sanitization: The skill uses defusedxml for XML parsing in all major processing scripts, which mitigates standard XML-based attacks like XXE.

docx