القائمة
Use Crater CLI job commands to list, inspect, create, stop, and snapshot jobs.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Use Crater CLI admin job commands to list, delete, lock, keep, and clean jobs.
Crater CLI 管理员审批工单域：通过 crater admin order ... 查看、批准、拒绝、清理审批工单。仅在用户明确要求管理员审核时使用。
Crater CLI 用户审批工单域：提交、编辑、取消、查看当前用户的审批工单。管理员审核动作必须使用 crater-cli-admin-approval-order。
Crater CLI 用户视图读取域：指导 AI Agent 通过 crater node、job、image、account、resource、dataset、model-download、pod 等用户可见命令查看平台只读信息。管理员视图请使用 crater-cli-admin-read。
Crater CLI 共享基础：安全调用 crater 命令的通用规则，包括可执行文件选择、全局选项、--json、--no-interactive、--help、错误输出、退出码、错误场景判断、敏感信息处理，以及执行会修改用户环境的命令前的确认规则。处理任何 Crater CLI 操作前使用。
Crater 代码开发：在 backend/、frontend/、cli/ 下开发 Go 后端、React 前端、CLI、API、作业模板、组件、表单、hooks、i18n 与测试。用户修改 backend/、frontend/、cli/ 代码或前后端/CLI 联动时使用；调用 crater 命令则用 cli/skills/crater-cli-*；开始前须应用 crater-devel-shared。
| name | crater-cli-job |
| version | 0.1.1 |
| description | Use Crater CLI job commands to list, inspect, create, stop, and snapshot jobs. |
| metadata | {"requires":{"bins":["crater"]},"cliHelp":"crater job --help"} |
Use this skill when the user asks to operate Crater jobs from the CLI: list jobs, inspect details, view pods/events/YAML/templates, get Jupyter/WebIDE access, open SSH, create jobs, stop/delete jobs, or snapshot jobs.
CRITICAL — Before doing anything else, MUST read crater-cli-shared (possible path: ../crater-cli-shared/SKILL.md) for global options, non-interactive use, errors, confirmation, and secret handling.
crater job lscrater job get|pods|events|yaml|template <jobName>crater job token <jobName>, crater job secret <jobName>, crater job ssh <jobName>crater job snapshot <jobName>, crater job alert <jobName>, crater job delete <jobName>crater job create jupyter|webide ...crater job create custom ...crater job create tensorflow|pytorch --file request.jsonUse crater job ls --json --no-interactive before destructive actions to confirm the exact jobName. User-facing display names are not always accepted by job APIs.
For create commands, validate resource values before calling the platform. CPU, memory, and GPU counts must not be negative; task replicas must be positive. Workspace mounts use subPath:mountPath; dataset mounts use datasetID:mountPath; forwards use name:port.
For Jupyter/WebIDE access commands, the returned token or password is sensitive. Prefer JSON only when the next tool needs structured fields, and avoid echoing secrets into logs or issue bodies.
List running GPU jobs for a user:
crater job ls --user alice --status Running --json --no-interactive
Inspect a job:
crater job get jpt-alice-abcde --json --no-interactive
crater job pods jpt-alice-abcde --json --no-interactive
crater job events jpt-alice-abcde --json --no-interactive
Create a Jupyter job:
crater job create jupyter \
--name experiment-notebook \
--image harbor.example/project/jupyter:latest \
--cpu 4 \
--memory 16Gi \
--gpu 1 \
--gpu-resource nvidia.com/gpu \
--json --no-interactive
Create a distributed PyTorch job from an exact backend-compatible request:
crater job create pytorch --file pytorch-job.json --json --no-interactive
Stop or delete a job:
crater job delete jpt-alice-abcde --yes --json --no-interactive
crater job create tensorflow|pytorch intentionally uses --file because the backend accepts a nested tasks[] request. The CLI rejects unknown JSON fields. Keep the JSON aligned with the backend DTO fields: name, tasks, resource, image.imageLink, volumeMounts, envs, selectors, alertEnabled, template, and optional scheduling fields. Distributed TensorFlow and PyTorch jobs do not support backfill scheduling.