The Agent Skills Directory

[DYNAMIC_EXECUTION]: The recalc.py script automatically configures a LibreOffice macro by writing a StarBasic script (Module1.xba) to the local application configuration directory (~/Library/Application Support/LibreOffice/ or ~/.config/libreoffice/). This macro is subsequently executed via the soffice binary to perform headless workbook recalculations. This process modifies the user's host environment configuration and executes dynamically generated script content.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from .xlsx files using pandas and openpyxl as part of its data analysis and style validation workflows. If the agent is instructed to analyze a file provided by an untrusted source, instructions embedded in the cell data could potentially influence the agent's behavior.
Ingestion points: File reading occurs in build_xlsx.py (via load_workbook), validate_xlsx.py (via load_workbook), and recommended pandas workflows in SKILL.md (pd.read_excel).
Boundary markers: Absent. No explicit delimiters or instructions are used to separate cell data from the agent's instructions during processing.
Capability inventory: The skill has capabilities for file writing (openpyxl), script chaining via subprocesses (recalc.py, validate_xlsx.py), and environment modification through macro setup.
Sanitization: While validate_xlsx.py checks for 'forbidden terms', this is a quality-focused style filter rather than a security sanitizer for LLM prompts.
[COMMAND_EXECUTION]: The skill utilizes subprocess.run() across build_xlsx.py, validate_xlsx.py, and recalc.py to chain the execution of internal Python scripts and the soffice (LibreOffice) binary. While these calls do not use shell=True, they execute binaries and scripts based on provided file paths to handle formula evaluation and linting.

xlsx