The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several internal Python scripts that invoke subprocess.run to perform local utility operations. For example, xlsx_insert_row.py executes a companion script to update cell references, and libreoffice_recalc.py invokes a headless LibreOffice binary for dynamic formula recalculation. These executions are strictly scoped to local file manipulation and do not present an arbitrary command injection risk.
[INDIRECT_PROMPT_INJECTION]: The skill handles untrusted external spreadsheet data (Excel and CSV files) through xlsx_reader.py and pandas. While this creates an ingestion point for potential data-based injections, the capabilities provided to the agent are focused on structured file processing and local validation.
Ingestion points: Untrusted data enters the environment through xlsx_reader.py during structural discovery and pandas-based analysis of user-provided spreadsheets.
Boundary markers: The instructions do not specify specific delimiters for data interpolation, but the processing is handled through structured DataFrames.
Capability inventory: The agent has access to file system read/write operations and local execution of included Python utility scripts.
Sanitization: The xlsx_unpack.py script includes deterministic sanitization of ZIP archive member paths to prevent path traversal (Zip-Slip) attacks during file extraction.

minimax-xlsx