The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run in libreoffice_recalc.py and xlsx_insert_row.py to execute helper scripts and the LibreOffice binary. These calls are implemented using argument lists rather than shell strings, which significantly reduces the risk of command injection from user-provided file names or parameters.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and modify untrusted Excel and CSV files, which constitutes a potential attack surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent context through xlsx_reader.py (which uses pandas) and xlsx_unpack.py (which extracts the spreadsheet's XML structure from a ZIP archive).
Boundary markers: The instructions focus on technical XML and data manipulation. While it treats data as structured content (XML nodes), it does not explicitly define boundaries to prevent the agent from following instructions embedded within spreadsheet cells.
Capability inventory: The skill includes capabilities to write files, extract ZIP archives, and execute local Python scripts or the LibreOffice binary via subprocesses.
Sanitization: The skill uses standard XML parsing (ElementTree) and ZIP libraries. Notably, xlsx_unpack.py includes a robust sanitization check to prevent Zip-Slip (path traversal) attacks by verifying that extracted member paths remain within the target directory.
[SAFE]: The implementation of path traversal protection in xlsx_unpack.py follows security best practices for handling untrusted ZIP archives. The skill also avoids destructive operations like openpyxl's save() on existing files to prevent the accidental loss of advanced features like macros and pivot tables.

minimax-xlsx