The Agent Skills Directory

[COMMAND_EXECUTION]: Runtime C compilation and process injection in scripts/office/soffice.py. The script writes C source code to a temporary file, invokes gcc to compile it into a shared library, and uses the LD_PRELOAD environment variable to inject the library into the soffice (LibreOffice) process. This is designed as a technical workaround for environments where AF_UNIX sockets are blocked.\n- [COMMAND_EXECUTION]: Automated application configuration and macro execution in scripts/recalc.py. The script automatically creates a StarBasic macro (Module1.xba) in the user's LibreOffice configuration directory and executes it to recalculate formulas in Excel workbooks.\n- [PROMPT_INJECTION]: Surface for indirect prompt injection via untrusted spreadsheet data. The skill ingests external data from .xlsx, .xlsm, .csv, and .tsv files using pandas and openpyxl. Since the agent context includes file system access and subprocess execution (soffice, gcc), malicious instructions embedded in spreadsheet content could potentially influence agent behavior if not properly sanitized.

xlsx