The Agent Skills Directory

[COMMAND_EXECUTION]: The recalc.py script executes the soffice (LibreOffice) binary via subprocess.run to perform headless spreadsheet recalculations. The execution is handled through a list of arguments, mitigating basic command injection, and targets a well-known local utility.
[PERSISTENCE]: The skill establishes a persistent configuration by writing a StarBasic macro to the user's local LibreOffice directory. This is used to automate the recalculation process across sessions. Evidence: The setup_libreoffice_macro function in recalc.py creates the necessary directory structure and writes Module1.xba to the standard LibreOffice configuration path.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a surface for indirect prompt injection as it is designed to read and process untrusted data from external spreadsheet files. \n
Ingestion points: Content is ingested via pd.read_excel() and load_workbook() as seen in SKILL.md and recalc.py.\n
Boundary markers: While the skill provides guidelines for formatting, it lacks explicit instructions to the agent to disregard instructions embedded within the spreadsheet data itself.\n
Capability inventory: The agent has access to file-writing capabilities and the recalc.py script for command execution.\n
Sanitization: No explicit content validation or sanitization is implemented to check cell values for malicious instructions before processing.

xlsx