If a word is typed as a command, and there is no built-in command with that name, the directories in the sourcepath list are searched in order for a file with the name given by the word. If it is found, it is read in as a command file (as if it were sourced). Before it is read, however, the variables argc and argv are set to the number of words following the file-name on the command line, and a list of those words respectively. After the file is finished, these variables are unset. Note that if a command file calls another, it must save its argv and argc since they are altered. Also, command files may not be re-entrant since there are no local variables. Of course, the procedures may explicitly manipulate a stack.... This way one can write scripts analogous to shell scripts for ngnutmeg and ngspice.

Note that for the script to work with ngspice, it must begin with a blank line (or whatever else, since it is thrown away) and then a line with .control on it. This is an unfortunate result of the source command being used for both circuit input and command file execution. Note also that this allows the user to merely type the name of a circuit file as a command and it is automatically run. The commands are executed immediately, without running any analyses that may be specified in the circuit (to execute the analyses before the script executes, include a run command in the script).

There are various command scripts installed in /usr/local/lib/spice/scripts (or whatever the path is on your machine), and the default sourcepath includes this directory, so you can use these command files (almost) like built-in commands.