Well I think you well may be scuppered then.
The next best thing I can think of is....
any job wanting to do an OS command writes a file on the server using utl_file to a certian fixed dir, it writes the exact syntax of what it wants doing.
A Unix shell script scans this dir looking for files, if it finds one it does it and then deletes the file. then scans again or maybe sleeps a bit first would be better...
If you wanted the DB app to be able to acknowledge that the command has been performed then it would have to use utl file to poll for the files existence as soon as it is gone then it could assume the job is done.
This is well dodgy though, I'm sure someone else has a better solution....