菜单
Find a dataset, checkpoint, or file across every lab server — locations, sizes, duplicates. Use for "where is coco", "which server has X", or locating checkpoints and project directories.
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 | where-is |
| description | Find a dataset, checkpoint, or file across every lab server — locations, sizes, duplicates. Use for "where is coco", "which server has X", or locating checkpoints and project directories. |
Input: a name or pattern. Search cheap-to-expensive; stop once found unless the user wants every copy.
list_datasets per reachable host — scan the top-level report for the name
first; this covers most datasets with zero extra cost.
Not there → targeted find on the data roots, capped so slow NFS can't hang:
for r in /data* /scratch "$HOME"/data*; do timeout 30 find "$r" -maxdepth 3 -iname '*<name>*' 2>/dev/null | head -20; done
Checkpoints and code: also try -maxdepth 5 under likely project
directories learned from list_sessions cwds.
For each hit, get size and freshness (timeout-capped):
ls -ldh <path>; timeout 60 du -sh <path> 2>/dev/null
Table: host · path · size · modified. Note duplicates across hosts. If the
user is deciding where to run something, also note which of the hosts holding
the data currently has free GPUs (list_hosts) — data gravity. If nothing was
found, say exactly where you looked.