The Agent Skills Directory

[COMMAND_EXECUTION]: The script xlsx_recalc.py uses subprocess.run to invoke the soffice (LibreOffice) binary for headless formula recalculation. This execution is performed securely by isolating the HOME environment variable to a temporary directory and restricting the PATH during the call.
[DATA_EXPOSURE]: No hardcoded credentials or sensitive file paths (e.g., .ssh, .aws, .env) are accessed. The skill operates strictly on user-provided spreadsheet and CSV files.
[REMOTE_CODE_EXECUTION]: The skill does not download or execute remote scripts. It relies on the well-known and trusted openpyxl Python package for core functionality.
[PROMPT_INJECTION]: The instructions in SKILL.md are focused on productivity and do not contain patterns for bypassing safety filters, overriding agent behavior, or extracting system prompts.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests untrusted external data.
Ingestion points: Untrusted data enters the agent context through xlsx_read.py, xlsx_to_csv.py, and csv_to_xlsx.py.
Boundary markers: None are present; the skill treats cell content as raw data.
Capability inventory: The skill has the capability to write to the file system and execute soffice via subprocess.
Sanitization: No sanitization or escaping is performed on cell content before processing.
This attack surface is standard for data-processing utilities and does not represent a malicious finding.

xlsx