The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/office/soffice.py performs runtime compilation of hardcoded C source code using gcc. The resulting shared library is then injected into the soffice process via the LD_PRELOAD environment variable to intercept and redirect UNIX domain socket calls, bypassing environment-specific restrictions.\n- [COMMAND_EXECUTION]: The script scripts/accept_changes.py dynamically creates a LibreOffice StarBasic macro and installs it into a temporary user profile. This macro is subsequently executed by soffice to automate document processing tasks.\n- [COMMAND_EXECUTION]: Several components, including scripts/office/soffice.py, scripts/accept_changes.py, and ooxml/scripts/pack.py, utilize the subprocess module to execute external binaries such as gcc and soffice.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its processing of untrusted OOXML data from Word documents.\n
Ingestion points: The Document library in scripts/document.py and the unpack utility in scripts/office/unpack.py read and parse XML data from external files.\n
Boundary markers: The skill does not employ delimiters or explicit instructions to ignore embedded commands within the parsed content.\n
Capability inventory: The skill has extensive filesystem access and the ability to execute shell commands and compile code, which increases the potential impact of successful injection.\n
Sanitization: Although the implementation uses defusedxml to mitigate XXE vulnerabilities, the document content itself is not sanitized against instructional text before being processed by the agent.\n- [EXTERNAL_DOWNLOADS]: The skill documentation (docx-js.md and SKILL.md) references the requirement for the external Node.js package docx, suggesting installation via npm install -g docx.

docx