The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses runtime code generation and process injection techniques to overcome environment-specific limitations for LibreOffice.
The script scripts/office/soffice.py generates C source code at runtime and compiles it into a shared library using gcc. It then launches soffice with the LD_PRELOAD environment variable to inject this library, shimming standard system calls related to Unix domain sockets.
The script scripts/accept_changes.py dynamically generates a LibreOffice Basic macro (.xba file) and writes it to the temporary user profile to automate document editing tasks.
Multiple components (e.g., pack.py, unpack.py, accept_changes.py) utilize subprocess.run to execute external binaries including soffice, gcc, git, pandoc, and pdftoppm for document processing and conversion.
[EXTERNAL_DOWNLOADS]: The skill requires external software packages to be installed by the user to achieve full functionality.
The SKILL.md documentation instructs the user to install the docx library globally using npm install -g docx.
[PROMPT_INJECTION]: The skill processes untrusted content from Word documents, creating an attack surface for indirect prompt injection.
Ingestion points: The skill extracts text and XML data from .docx files using pandoc and custom Python parsing logic in scripts/office/unpack.py.
Boundary markers: No explicit delimiters or instructions are provided to the agent to distinguish between its own logic and the processed document content.
Capability inventory: The skill possesses significant capabilities, including file system access, runtime compilation, and the execution of powerful binaries like soffice and pandoc.
Sanitization: While the skill uses defusedxml to mitigate XML-based attacks (e.g., XXE), it does not implement sanitization or filtering for natural language instructions embedded within the document text.

docx