The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill performs runtime compilation and library injection to facilitate its operations.
scripts/office/soffice.py contains a hardcoded C source string (_SHIM_SOURCE) that is written to a temporary file and compiled using gcc -shared -fPIC at runtime.
The resulting shared library is loaded via the LD_PRELOAD environment variable to intercept socket-related system calls during LibreOffice execution.
scripts/accept_changes.py dynamically generates a LibreOffice Basic macro (Module1.xba) in a temporary profile directory to automate the acceptance of tracked changes.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted documents from external parties, creating an indirect prompt injection surface.
Ingestion points: The skill reads and processes content from .docx and .dotx files using pandoc, unzip, and XML parsers.
Boundary markers: There are no explicit delimiters or instructions to the agent to isolate document text from the instruction context.
Capability inventory: The skill executes several powerful system utilities (soffice, pandoc, pdftoppm, gcc, zip, unzip) via subprocess.run.
Sanitization: The skill mitigates some archive-based attacks by removing symlinks from unpacked directories but does not sanitize text content for embedded instructions.
[COMMAND_EXECUTION]: The skill makes extensive use of subprocess.run to interact with the operating system and execute external binaries for document processing and compilation.

docx