The Agent Skills Directory

[COMMAND_EXECUTION]: The recalc.py script utilizes the subprocess.run function to execute system commands, specifically calling the soffice (LibreOffice) binary and the timeout utility. These commands are executed as an argument list, which is a secure practice to prevent shell injection, but they involve direct interaction with the host operating system to facilitate formula recalculation. Additionally, the script writes a StarBasic macro to the application configuration directory (e.g., ~/.config/libreoffice) to automate the recalculation process.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external spreadsheet files (XLSX, CSV, TSV) using the pandas and openpyxl libraries. Because it processes untrusted data which is then used to drive subsequent agent actions (such as formula creation or model modification), it presents a surface for indirect prompt injection.\n
Ingestion points: Data is loaded into the environment from spreadsheet files via pd.read_excel and load_workbook (SKILL.md).\n
Boundary markers: The instructions do not specify the use of delimiters or 'ignore embedded instructions' warnings when processing cell content.\n
Capability inventory: The skill has the ability to write to the filesystem, execute subprocesses (via recalc.py), and format data across workbooks.\n
Sanitization: There are no documented procedures for sanitizing, filtering, or validating the content of the cells before they are processed by the agent.

xlsx