What problem does it solve? Login-node Jupyter kernels on shared HPC systems like RCC Midway are silently OOM-killed because the memory ceiling applies to the whole user across all sessions, causing lost variables and reset cell counters. This Skill moves kernels onto compute nodes with dedicated memory and provides recovery workflows when kernels die. ## Core Features & Use Cases - Compute-node kernel startup: Submit SLURM sbatch jobs that launch ipykernel bound to the node's routable IP, with partition and account handling for RCC Midway and UMich Great Lakes. - Kernel lifecycle management: Detect stale kernels, kill old SLURM jobs, connect via connection files, and verify hostname and Python executable. - Wall-time autosave and recovery: Automatically snapshot notebook state every 30 minutes and restore sessions into a fresh kernel after a job ends or a kernel dies. - Use Case: A pandas kernel on the Midway login node is OOM-killed mid-analysis. Use this Skill to submit a 48G caslake compute job, connect to its kernel, restore the autosaved session, and continue the notebook without re-running expensive steps. ## Quick Start Connect to a compute kernel on Midway with 48G of memory so my notebook stops getting OOM-killed on the login node.
