I’m evaluating an offline command-line tool launched from Terminal on macOS. The parent process creates an enclosing directory, then launches the tool through /usr/bin/sandbox-exec -p with an inline profile. The tool must create its own output directory, which is absent at launch.
Can a macOS sandbox profile reliably enforce all of the following for that invocation?
- Allow reads of the tool’s interpreter, dependencies and input files.
- Deny network access and creation of child processes by the sandboxed tool.
- Allow the tool to create
/path/to/work/proof/producer001and write beneath it, while denying writes outside that directory, including writes through symlinks or other path aliases.
The proposed profile denies file-write* globally and allows it only under a subpath rule for the output directory. Does that exception cover creation of the directory itself when only its parent exists? Does deny process-fork cover all child-process creation paths, including posix_spawn?
Is there a supported way to inspect or validate the effective policy for an exact command and working directory before invoking it? If sandbox-exec is unsuitable for this use, is there a supported macOS mechanism for applying these restrictions to a command-line tool launched from Terminal? A documentation reference would be helpful.