The Agent Skills Directory

[DYNAMIC_EXECUTION]: The file scripts/office/soffice.py dynamically generates C source code at runtime, writes it to a temporary location, and compiles it into a shared library using gcc. It subsequently uses the LD_PRELOAD environment variable to inject this library into the LibreOffice (soffice) process. This mechanism is documented as a workaround to allow LibreOffice to function in restricted environments by intercepting and redirecting Unix socket system calls to use socket pairs.
[COMMAND_EXECUTION]: The skill uses subprocess.run across several scripts to execute external system tools. This includes invoking gcc to compile technical shims in scripts/office/soffice.py, running soffice for model recalculation in scripts/recalc.py, using the file utility for document format identification in scripts/parse_complex_excel.py, and using git diff for redlining validation in scripts/office/validators/redlining.py.
[INDIRECT_PROMPT_INJECTION]: As the skill is designed to ingest and process untrusted Office documents provided by users, it inherently possesses a vulnerability surface for indirect prompt injection. While the skill employs defusedxml to protect against XML-based attacks like XXE, the automated extraction and processing of content from complex binary formats like Excel and Word could be used to deliver malicious instructions into the agent's context.

xlsx