docx
Warn
Audited by Gen Agent Trust Hub on Jun 16, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill performs runtime C compilation and process injection in
scripts/office/soffice.py. It writes C source code for a socket-shimming library to a temporary location, compiles it usinggcc, and then uses theLD_PRELOADenvironment variable to inject the resulting library into thesoffice(LibreOffice) process. This logic is intended to redirect AF_UNIX socket calls tosocketpairto enable LibreOffice in environments with socket restrictions. - [REMOTE_CODE_EXECUTION]: In
scripts/accept_changes.py, the skill dynamically generates a StarBasic macro and writes it to the filesystem inside a temporary LibreOffice profile. It then invokessofficeto execute this macro, enabling automated acceptance of tracked changes in headless mode. - [EXTERNAL_DOWNLOADS]: The
SKILL.mdinstructions guide the installation of thedocxpackage from the public NPM registry to handle document generation tasks. - [COMMAND_EXECUTION]: Multiple scripts utilize the
subprocessmodule to execute external binaries includingsoffice,pandoc,git, andgcc. These commands are used for file conversion, extraction, diffing, and shimming. - [COMMAND_EXECUTION]: The skill processes untrusted document data and maintains significant capabilities including file system access and shell command execution. This configuration represents an attack surface where maliciously crafted documents could attempt to influence agent actions.
- Ingestion points:
scripts/office/unpack.py(reads and extracts XML content from user-supplied .docx files). - Boundary markers: Absent; the skill instructions rely on direct XML manipulation and string interpolation without explicit boundary protection for external content.
- Capability inventory: Runtime compilation via
gcc, process shimming viaLD_PRELOAD, and command execution for document conversion and analysis inscripts/office/soffice.py,scripts/accept_changes.py, andscripts/office/validators/redlining.py. - Sanitization: Employs
defusedxml.minidomfor XML parsing, which provides protection against XML External Entity (XXE) attacks.
Audit Metadata