Single-workstation GPU job queue helper using tsp/task-spooler. Use for non-Slurm background training, sequential local runs, single-machine GPU queueing, or "줄 세우기" on a shared box.
Instalación
Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.
Single-workstation GPU job queue helper using tsp/task-spooler. Use for non-Slurm background training, sequential local runs, single-machine GPU queueing, or "줄 세우기" on a shared box.
Default: one local GPU queue, sequential unless the user asks otherwise.
Rules
Prefer the queue over launching many GPU jobs at once on a shared workstation.
Default to --slots 1 to avoid OOM; raise slots only when resource use is clear.
Never put secrets on the command line or in ledger notes; pass them via environment or files.
Use ~/.oh-my-setting/scripts/tsp-queue.sh as the entrypoint. It records completions to the run ledger.
Cluster jobs go through slurm-hpc; single-box jobs go through tsp-queue.
