Token导航 LogoToken导航

Agent Skill · 效率

sjtu-slurm-skill(上海交通大学泥浆技能)

sjtu-slurm-skill 用于补充效率相关能力,适合在 OpenClaw 中需要让 Agent 承接效率相关任务时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

sjtu-slurm-skill 是效率类 Agent Skill,可用宿主为 OpenClaw,通过ClawHub获取,权限为接触敏感数据;许可证为 MIT-0,第三方扫描已通过,按安装量在效率类的 21,003 个 Skill 里排第 8,885。

总安装 6,610周安装 270GitHub Stars 公开资料未说明下载 2,138核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
MIT-0
版本
0.2.3

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:sjtu-slurm-skill(上海交通大学泥浆技能)
来源仓库:https://github.com/taleintervenor/sjtu-slurm-skill
安装命令:
openclaw skills install sjtu-slurm-skill
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 OpenClaw 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

ClawHubOpenClaw
openclaw skills install sjtu-slurm-skill

这个 Skill 是做什么的

效率类 Skill 把重复的整理、汇总、转换工作交给助手完成,减少手工复制粘贴。

以用户身份登录交大 HPC 平台“交我算”,进行作业管理与数据操作。支持作业查询、提交、取消及数据迁移等任务。适用于高性能计算场景下的资源调度和作业监控。安装前建议确认账号权限和配额使用情况。sjtu-slurm-skill 属于效率类 Skill,可作为该场景下的辅助能力补充。

注意维护状态,避免因平台策略调整影响调用。

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 OpenClaw;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 openclaw skills install sjtu-slurm-skill;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要接触敏感数据,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

sjtu-slurm-skill 怎么安装?

在终端执行 openclaw skills install sjtu-slurm-skill。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

sjtu-slurm-skill 支持哪些 AI 工具?

公开资料标注支持 OpenClaw。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

sjtu-slurm-skill 需要什么权限?

权限级别是接触敏感数据。该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

sjtu-slurm-skill 是免费的吗?

来源标注的许可证是 MIT-0。本站不代收任何费用,也不托管安装包;能否商用、要不要署名,以许可证原文为准。

sjtu-slurm-skill 用的人多吗?

公开数据是累计安装 6,610 次,近一周安装 270 次,下载 2,138 次。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

sjtu-slurm-skill 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

值得选的地方和要先想清楚的地方

值得选的地方

  • 专注 OpenClaw 一种宿主,触发方式和使用习惯不用来回切换。
  • 许可证明确为 MIT-0,商用和分发边界好判断。
  • 第三方扫描已通过,常见的高风险行为没有命中。

先想清楚的地方

  • 需要接触敏感数据,装到常用环境前要先评估影响面。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 OpenClaw,想直接扩展它能力的人。
  • 需要处理效率类任务、又不想每次都手写提示词的人。
  • 需要许可证清楚、并且有第三方扫描结果的团队项目。

不太适合

  • 环境里有大量密钥、凭证,不方便给第三方 Skill 读取的人。
  • 需要厂商级技术支持和 SLA 的生产系统。

效率类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
github

GitHub 代码协作 · OpenClaw

7,632,918545需要联网MIT-0通过
nano-pdf

Nano PDF 文档 · OpenClaw

4,249,476231需要联网MIT-0通过
auto-updater

自动更新程序 · OpenClaw

2,964,063386操作浏览器MIT-0通过
video-frames

视频帧 · OpenClaw

2,621,304112需要联网MIT-0通过
gemini

Gemini 助手 · OpenClaw

2,547,39648需要联网MIT-0有提醒
sjtu-slurm-skill

上海交通大学泥浆技能 · 本页

6,610—接触敏感数据MIT-0通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

name
sjtu-slurm-skill
description
Log in to the SJTU HPC platform (also known as "交我算") as the user to perform job queries, submissions, cancellations, and data management. Use this skill when the user requests operations related to HPC or "交我算".
metadata

sjtu-slurm-skill

Overview

Use this skill to log in to the SJTU HPC (交我算) platform as the user, acting on their behalf to perform personal job queries, submissions, cancellations, and data management operations.

General Principles:

  • Risk-aware operations: Deleting user's data and interrupting running jobs are risky operations. Before performing any risky operation, always confirm with the user that they clearly understand the impact of the operation and agree to its execution.
  • private credentials: All user's credential files (like SSH key, certificate, API token, etc...) should be stored in the credentials directory under workspace (e.g., ~/.openclaw/workspace/default/credentials). Do Not store user's plain text password or contain it in your talk messages.

Quick Start

  1. Ensure HPC API token file is available in the workspace, which should be stored in credentials directory. If not, request a new token.
  2. For each user request, analyze whether you need to log in to an HPC entry node to perform the operation remotely. You can ask more questions to clarify any ambiguous parts of the request.

- If user wants to know its storage quota usage or update its account (like password, binding Email/jAccount, preferred contact method), then you can meet the requirement directly by calling HPC API with the token. - if user is talking about job or its data, then you have to select an entry node to do remote operation. In this case, take the following steps. - if user want to get a passwordless certificate for SSH login, refer to certificates section.

  1. Ensure SSH keys and certificates are available in the workspace, which should be stored in credentials directory. If not, request a new SSH certificate for the user. Remind the user that requesting a certificate will trigger two-factor authentication.
  2. For each user request need remote operation, identify the node group/partition the user is interested in and the operation type to select the correct entry node.
  3. Use the SSH certificate to connect to the corresponding entry node based on the cluster and operation type, execute the user's requested operations on it.

Ensure Token Available

Before calling any HPC API, ensure there is a valid token in the credentials directory. Go through the following steps:

  1. Check whether hpc_token file exists in the credentials directory under workspace. If it exists, goto step 2, otherwise goto step 3.
  2. Try to refresh the token with refresh_token.py --workspace "path_to_workspace". If success, then you have ensured a valid token and can safely skip the following steps. If the request is refused (which means the token has expired), then continue to step 3. If the request reports internal error, act according to the rules in Error Handling section .
  3. Tell the user you are going to request a new token and ask for their HPC username and password. To avoid password leak, advise user to upload a text file instead of providing passwords directly in the chat window.
  4. Call script to request a new token with req_token.py :

- If user uploaded file to provide password, call script with the format like: req_token.py "username" "password" --workspace "path_to_workspace" --textfile "uploaded_file_path". So script can help to remove the file to prevent password leak. - Otherwise skip the --textfile option.

The token will be saved as hpc_token in the credentials directory under workspace. If any script failed, act according to the rules in Error Handling section .

HPC API

Here are some HPC API path you can call when users want to know their storage quota usage or update their account:

  • GET /user: query user's information.

- Response will includ all the attributes defined by posixAccount and some addditional fields like jAccount, email, user type, etc. - CURL example: curl -L 'https://api.hpc.sjtu.edu.cn/user?name=userA&domain=pi' --header 'Authorization: Bearer ***'

  • PATCH /user: update user's information.

- Only password, Email/jAccount, preferred contact method can be updated by user itself. - Update request will trigger two-factor authentication via JWB APP (交我办) or Email. Ask user whether it have been prepared to do the authentication before you send the request. - Consult the online documentation to understand the calling conventions of this API.

  • GET /quota: Query the storage quota data of user or its related group account.

- Call this API does not need authorization. - Always call it with explict name parameter. - If user's requirement didn't specify which quota_type them need, you should query with quota_type=account and quota_type=user seperately. The account name can be derived from user's home path, which can be query in GET /user. The parent dir of user‘s home is exactly its account name, for example: "home": "%H/home/acct-hpc/hpcrobot" -> account name: acct-hpc. - CURL example: curl -L 'https://api.hpc.sjtu.edu.cn/quota?quota_type=account&name=acct-example'

Only these tasks can be done through HPC API. Do NOT try to use API for any other user requirements.

Use the token in credentials directory if authorization is required. Ask for user's HPC name if it is needed and has not been provided during the talk.

Entry Node Selection Rules

Select the appropriate entry node based on the node group and operation type the user is interested in.

HPC cluster has 3 node groups: pi, sy (思源) , kp (鲲鹏) . Partitions belong to node groups. If user does not explicitly specify a node group but only provides a target partition or a target storage, look up the corresponding node group from this table:

node grouppartitionstorage
picpu, debug, dgx2, huge, 192c6tlustre
kpdebugarm, arm128c256g, scnet_arm
sysmall, 64c512g, el9, debug64c512g, win32, a100, a800, debug100dssg(gpfs)

Then select the entry node according to this table:

businessnode groupentry node
jobpipilogin.hpc.sjtu.edu.cn
jobsysylogin.hpc.sjtu.edu.cn
jobkparmlogin.hpc.sjtu.edu.cn
datapi, kpdata.hpc.sjtu.edu.cn
datasysydata.hpc.sjtu.edu.cn

SSH Keys and Certificates

SSH login to entry nodes requires the user's passwordless certificate. The agent should store all credentials (tokens, SSH keys, certificates) in the workspace's credentials directory. If no certificate available, follow the steps to request a new SSH certificate:

  1. Ask for certificate owner: If you have asked username when requesting token, directly use that username and skip the ask. Otherwise ask for user's HPC username.
  2. Make sure user is prepared: Tell the user that requesting the certificate will trigger two-factor authentication via JWB APP (交我办) or Email. Ask user whether it have been prepared to do the authentication.
  3. Call synchronous script with long timeout: Give the user a prompt like "Requesting certificate now, please check your JWB App (交我办) or Email...", Do NOT wait for the user's confirmation before proceeding. Immediately call the script req_certificate.py <username> --workspace "path_to_workspace" with timeout >= 600s because the script will block waiting for the user to complete two-factor authentication on another channel.
  4. Handle errors: After script execution, check the exit code. If non-zero, act according to the rules in Error Handling section .

The SSH key and certificate will be saved to the credentials directory under workspace.

If user directly asked for the key or certificate, follow the steps to send files to user:

  1. Check whether the valid time of the existing certificate meets the user's requirement. If not, request a new certificate with --valid-time argument.
  2. Check if there is any channel tools or skills can be used to send files directly to user. List all the candidate methods and ask user which one is preferred.
  3. Use the selected method to send private key and certificate file to user. Remind user of the security sensitivity of these files and advise user never use them in public environments.

Error Handling

If any script execution or API call failed, parse the error message from stderr and inform the user with clear details:

  • If the error message indicates "user have not set email or jAccount", Direct the user to read the platform documentation: https://docs.hpc.sjtu.edu.cn/accounts/security.html . Ask them to follow the instructions in the documentation to bind their second identity channel (jAccount or email).
  • If the error message indicates internal error, tell user the session ID in error message, suggest them to ask help from hpc@sjtu.edu.cn with this ID.

Executing User-Requested Operations on Entry Nodes

Use the SSH key and certificate from the workspace credentials directory to log in to the entry node and remotely execute the user's requested operations:

ssh -i "/path_to_workspace/credentials/private_key" -o "CertificateFile=/path_to_workspace/credentials/certificate" user@entry_node "command"

Note that user data is distributed across multiple shared storage pools, and visible storage varies depending on the node group. The storage where the user's home directory resides also differs:

node groupmount pointstorage information
pi, kp/lustreHot storage, Lustre, user home directory is here
sy/dssgHot storage, GPFS, user home directory is here
pi, sy, kp/archiveCold storage A, NFS, for archived data, writable only on data and sydata nodes, read-only on other nodes
pi, sy, kp/vaultCold storage B, NFS, for archived data, writable only on data and sydata nodes, read-only on other nodes
pi, sy, kp/unionmergerfs virtual filesystem, combines /archive and /vault, same read/write restrictions as the two cold storages

User personal directory paths are consistent across storage pools. Replace the top-level path of the user's home directory with the target storage to get the corresponding personal directory path. For example: /lustre/home/acct-hpc/hpcrobot -> /archive/home/acct-hpc/hpcrobot

If the user is interested in platform status or personal jobs, use SLURM commands to perform operations like: sinfo, squeue, sacct, seff, sbatch, scancel, etc...

Due to different storage pool visibility across node groups, job data computed on a partition can only be viewed on entry nodes belonging to the corresponding node group. Each entry node only allows submitting job to partitions within its own node group.

Never use cold storage data directly for job. If cold storage data is needed, transfer it to hot storage first before using it for job.

Large-scale data transfer operations should be performed on data.hpc.sjtu.edu.cn and sydata.hpc.sjtu.edu.cn. To aviod connection lost, wrap the remote transfer operation on data nodes with tmux. HPC has multiple cold storages, so before writing data to cold storage, check the free space of these candidates, and select the cold storage with most free space as the write target. If user asks about their data in cold storage and did not clearly specified which one, use the combined virtual filesystem /union .

Resources

scripts/

  • req_token.py: Request a new bearer token from HPC API. This is a one-time setup that enables other operations.
  • refresh_token.py: Refresh an existing bearer token to extend the session without re-authenticating.
  • req_certificate.py: Request SSH key and certificate files using an existing token, which can be used to log in to the entry nodes without a password.

平台分布

OpenClaw

70.8%
按下载量换算1,514

安全审计

VirusTotal

通过

ClawScan

通过

Static analysis

通过

权限和风险

接触敏感数据

该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills