The Agent Skills Directory

PROMPT_INJECTION (LOW): Detected surface for indirect prompt injection as the skill processes external Excel files.
Ingestion points: recalc.py reads user-provided Excel files using the openpyxl library.
Boundary markers: Absent. The script identifies and returns cell coordinates and error strings (e.g., #REF!) which are then passed to the agent context without specific encapsulation.
Capability inventory: The script can execute subprocesses (soffice) and read/write files in the local filesystem.
Sanitization: Input filenames are converted to absolute paths and passed as list arguments to subprocess.run, which prevents shell injection.
COMMAND_EXECUTION (SAFE): The script executes soffice and timeout using subprocess.run with a list of arguments. This is a best-practice implementation that prevents command injection.
Persistence (LOW): The setup_libreoffice_macro function writes a StarBasic macro to the user's LibreOffice configuration directory (Module1.xba). While this is a persistent modification of application configuration, it is the primary mechanism required for the skill's stated purpose and contains no malicious logic.

xlsx