The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves reading and editing external Excel files using pandas and openpyxl, which establishes an indirect prompt injection surface where malicious data in a cell could be interpreted as instructions by the agent.
Ingestion points: SKILL.md provides instructions for the agent to ingest data from external spreadsheets using pd.read_excel() and load_workbook().
Boundary markers: There are no explicit instructions or delimiters defined to help the agent isolate spreadsheet data from its own command instructions.
Capability inventory: The skill has the capability to write to the local file system and execute the soffice binary via the recalc.py script.
Sanitization: No sanitization or validation of data retrieved from cell contents is implemented in the provided scripts or instructions.
[COMMAND_EXECUTION]: The recalc.py script executes the LibreOffice binary (soffice) using subprocess.run to perform formula recalculation. The command is invoked safely using a list of arguments, which prevents shell-based injection attacks. The script also configures a local LibreOffice macro to support headless recalculation, which is a necessary step for the documented functionality.

xlsx