I wish it would run in a stricter sandboxed mode and prompt the user on the first network requests and file writes outside of it's directory.

That wouldn't be perfect, but at least the user could be prompted for a concrete action instead of a vague "this script is scary" warning.