메뉴
Submit, monitor, and manage SLURM compute jobs on the PVE cluster (ctl + node01). Covers sbatch, srun, squeue, scancel, sacct, and MiQi-specific job templates.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
将 GitHub 仓库的所有 Issue 同步到飞书多维表格。自动创建匹配字段（标题/编号/状态/链接/创建人/标签/创建时间/更新时间/正文），支持增量去重、dry-run 预览、批量写入。适用场景：用户提到同步 GitHub Issue 到飞书、把 Issue 写入多维表格、关联 GitHub 和飞书等需求。
Electron 34+ 桌面应用端到端测试技能。当需要为 Electron 应用编写或调试 E2E 测试、遇到 Playwright _electron.launch 不可用、CI 中 Chromium SIGTRAP 崩溃、ELECTRON_RUN_AS_NODE 导致 Electron 以 Node 模式运行、或需要 CDP 远程调试连接方案时使用此技能。涵盖 CDP 连接、进程管理、跨平台 kill、CI sandbox 处理、concurrency 策略。适用于 Electron 34 及以上版本。
生成实习生结构化工作记录 docx 文件。当用户说"日报"、"新增"、"周报"、"任务看板"、"台账"、"生成docx"、"写日报"、"填新增"时触发。支持日报、新增记录、周报、任务看板（五表联动）、工作台账（四表联动）五类文档的自动生成。输入为口语化的工作内容描述，输出为格式化的 .docx 文件。
Push, monitor, debug, and iterate on Kaggle kernel notebooks using the kaggle CLI. Covers: nbformat notebook generation, kernel-metadata.json configuration (GPU/dataset), push→poll→debug→fix→repush cycle, papermill timeout avoidance, and output download. Use when the user wants to run a notebook on Kaggle cloud GPUs and iteratively fix errors until it succeeds.
Use this skill when creating Python scripts that target Kaggle Notebooks. This skill ensures correct cell separator format (# %% [code] and # %% [markdown]), proper markdown cell formatting (each line prefixed with #, no triple quotes), and Kaggle-specific conventions (Chinese comments, English output, %pip install). Also covers common pitfalls: OpenCV video decoding limits and Chinese text rendering with PIL.
SkillHub 一站式技能管理。从 my-skills registry 搜索和安装技能（明文/加密）。加密技能支持本地安装（保存 .dat + 元数据），使用时解密。触发词：安装技能、搜索技能、find skill、skill install、解密技能、encrypted skill、skill decrypt。
| name | slurm |
| description | Submit, monitor, and manage SLURM compute jobs on the PVE cluster (ctl + node01). Covers sbatch, srun, squeue, scancel, sacct, and MiQi-specific job templates. |
Instructions for the agent to follow when this skill is activated.
Use this skill when the user wants to submit, monitor, cancel, or inspect jobs on the SLURM cluster.
| Host | Role | Access |
|---|---|---|
ctl | Control node — submit jobs | ssh ctl |
Windows 用户: SLURM 命令无法在本地运行，必须先 ssh ctl，再在 ctl 上执行。也可 ssh ctl "squeue" 直接远程执行。
上传脚本: scp job.sh ctl:~ | 取回日志: scp "ctl:~/slurm-%j.out" .
sinfo # cluster/partition status
squeue -u $USER # my jobs
sbatch job.sh # submit
srun --pty bash # interactive shell
scancel <jobid> # cancel job
scancel -u $USER # cancel all my jobs
sacct -j <jobid> --format=JobID,State,Elapsed,MaxRSS
scontrol show job <jobid> # job details
ssh ctl "sbatch ~/job.sh" # submit 不要使用转义 按照示例格式输入
ssh ctl "squeue -u $USER" # my jobs
ssh ctl "scancel <jobid>" # cancel job
ssh ctl "sacct -j <jobid>" # job details
不需要把脚本从 Windows 格式转成 Linux 格式
目前没有nfs共享文件系统，所以不能查看任务输出