菜单
Find zombie processes holding GPU VRAM on lab servers (dead notebook kernels, orphaned trainers) and propose kill commands. Report-only. Use when GPUs look occupied but idle, or for "who is using GPU N".
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Find what's eating disk on full lab servers — old checkpoints, wandb and cache junk — and produce a deletion plan. Report-only; deletion needs per-path human approval. Use when a disk is full or the user asks to free space.
Compare experiment results across servers — differing hyperparameters and final metrics from local wandb data, with dashboard links. Use for "which ablation won" or comparing runs of a project.
Morning briefing across all lab servers — what finished or crashed overnight, which GPUs are free, which disks are filling up. Use when the user asks for a briefing, "what happened overnight", or overall lab/server status.
Launch a training run on the best lab server — picks a host by data locality and free GPUs, verifies code and environment, starts it in a detached tmux session. Use when the user wants to run, launch, or train something on the servers.
Reproduce a working project environment from one lab server on another — code at the same commit, a CUDA-compatible python env, and a smoke test. Use for "make this run on host B too" or setting up a new server.
Check on the user's running experiments across all servers — progress, loss, ETA, stalled or crashed. Use for "how are my runs", "is training still going", or checking a specific experiment.
| name | zombie-hunt |
| description | Find zombie processes holding GPU VRAM on lab servers (dead notebook kernels, orphaned trainers) and propose kill commands. Report-only. Use when GPUs look occupied but idle, or for "who is using GPU N". |
Goal: explain every GPU whose memory is held, and flag holders that are probably dead weight. You never send a signal yourself.
list_hosts → hosts where GPUs are inUse. For each, via run_command:
Map memory to processes:
nvidia-smi --query-compute-apps=gpu_uuid,pid,process_name,used_memory --format=csv,noheader
nvidia-smi --query-gpu=index,uuid --format=csv,noheader
Inspect each PID:
ps -o pid,ppid,user,etime,stat,%cpu,cmd -p <pid1>,<pid2>,...
Zombie heuristics — call it a zombie only on 2+ signals (or 1 signal plus user confirmation):
T (stopped) or defunctOwnership matters. The user's own PIDs → provide kill <pid> (then
kill -9 as fallback) for the USER to run. Other users' PIDs → never
suggest killing; report "GPU 2 held by 's idle kernel (14h, 0% cpu) —
ping them".
Table: host · GPU · PID · owner · age · cmd (trimmed) · verdict (active / probably zombie / unknown) · suggested action. Include how much VRAM each verdict would free.