The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted Excel and CSV files through scripts like xlsx_reader.py (using pandas) and various XML surgery tools. This introduces an attack surface where malicious data could potentially influence agent behavior. However, the skill provides specific instructional boundaries and the tools are focused on structured data manipulation.
Ingestion points: xlsx_reader.py (pandas), xlsx_unpack.py (zip extraction).
Capability inventory: File writing (xlsx_pack.py), command execution (libreoffice_recalc.py, xlsx_insert_row.py), and file system access.
Sanitization: xlsx_unpack.py includes explicit protection against path traversal (Zip-Slip) attacks by validating that extracted members remain within the target directory.
[COMMAND_EXECUTION]: The skill utilizes subprocess.run in libreoffice_recalc.py to invoke LibreOffice for formula recalculation and in xlsx_insert_row.py to call sibling scripts. These calls use structured argument lists and are restricted to the skill's intended operational scope.
[SAFE]: The skill demonstrates security awareness by implementing Zip-Slip protection in its decompression logic and follows the principle of least privilege by using targeted XML edits rather than fully rewriting spreadsheets with non-transparent libraries.

minimax-xlsx