Right now if a job fails before agent starts up, we don't get any reason, just slurm failure. We should put the slurm job logs in a findable place.