RunTestScript saves the test script and executes it. The script is a Starlark program which calls add_test(...) once per test; its result replaces the tests the script produced last time: a test at the same testset and index is updated in place and keeps its id, a new one is created, one no longer produced is deleted. Tests created by hand are never touched, and an index they occupy is an error for the script. Everything is written as one problem version, so one generation run follows, as after any test change. An empty source is how the script is removed, and takes its tests with it. Pass dry_run to get the changes the run would make without saving or writing anything, which is how a script is previewed. A script error fails the call with its line and nothing is written.