-Send mail on success and/or failure.
+Allow the output of a job to be listed. This is pretty easy since we
+know the tmpdir and the output file.
How to deal with stuck / long-running jobs?
job "foo" timeout 1 hour ... # timeout the job after an hour
job "bar" single instance ... # only one instance can run
job "bar" kill previous ... # kill previous if still running
+
+Implement native dynlink.