The API for the shell function is kind of neat, in that it seems to prevent you from accidentally creating shell injection vulnerabilities by calling it without properly quoting the arguments.

For example, in Python you could easily do this:

  message = '; cat /etc/passwd'

  # Whoops, shell injection vulnerability!
  subprocess.run(f'echo Message: {message}', shell=True)

  # Correct (assuming sh-compatible shell).
  subprocess.run(f'echo Message: {shlex.quote(message)}', shell=True)

  # Correct (without using shell).
  subprocess.run(['/bin/echo', 'Message:', message])

But the Bun API doesn't separate quoting from executing the command, so you can't make that kind of mistake:

  let message = '; cat /etc/passwd';

  // Works correctly.
  await $`echo Message: ${message}`.text();

  // Fails safely by throwing error about incorrect usage.
  await $('echo Message: ' + message).text();