The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/office/soffice.py contains logic to write a C source file to the system's temporary directory, compile it into a shared library using gcc, and then use the LD_PRELOAD environment variable to inject this library into the soffice (LibreOffice) process. This is used to shim system calls related to AF_UNIX sockets, which may be blocked in some environments.
[COMMAND_EXECUTION]: The skill makes several calls to external system binaries using subprocess.run:
gcc: Used to compile the compatibility shim in scripts/office/soffice.py.
soffice: The primary LibreOffice binary used for formula recalculation in scripts/recalc.py and scripts/office/soffice.py.
git: Used for generating document diffs in scripts/office/validators/redlining.py.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it is designed to ingest and process untrusted spreadsheet data (XLSX, CSV, etc.) from users. This data is then interpreted by the agent, creating a surface where malicious instructions embedded in a spreadsheet could attempt to influence the agent's behavior. The skill includes some XML sanitization via the defusedxml library, which mitigates standard XML-based attacks.

xlsx