Been there, did that. Wrote shell utilities like that. No linking of Startup.obj, just my code and the linker that turned my .obj into a scatterloadable executable. Run it from the shell, the system provides a default stack and jumps right into the first function.

The whole executable was less than a kilobyte in size. It accessed the system tick timer, used it to seed a RNG, which was actually part of the executable, chose a random quote from a megabyte-sized fortune file, and printed it. Everything clean, all resources closed and returned. It even had error handling.