SkyPilot Multi-Cloud Orchestration
Orchestra-Research/AI-Research-SKILLs
Runs ML training and batch jobs across clouds with SkyPilot, using spot instances, automatic region selection and managed recovery to cut GPU cost.
Operates the Inspire ML platform through its local `inspire` CLI: picking account, workspace and resources, launching notebooks, jobs and services, then cleaning up.
SKILL.md written in Chinese; this summary is our English description.
$ npx skills add realZillionX/InspireSkill --skill inspire -a claude-codeProject install by default; add -g for ~/.claude/skills/.
$ gh skill install realZillionX/InspireSkill inspire --agent claude-codeProject scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).
Claude Code skills documentation · loads skills from .claude/skills/
Install the "inspire" agent skill from https://github.com/realZillionX/InspireSkill/tree/main into .claude/skills/inspire/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "inspire", then confirm the skill loads.Claude Code copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add realZillionX/InspireSkill --skill inspire -a codexProject install goes to .agents/skills/; add -g for ~/.codex/skills/.
$ gh skill install realZillionX/InspireSkill inspire --agent codexProject scope by default (.agents/skills/); add --scope user for a personal install.
Codex skills documentation · loads skills from .agents/skills/
Install the "inspire" agent skill from https://github.com/realZillionX/InspireSkill/tree/main into .agents/skills/inspire/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "inspire", then confirm the skill loads.Codex copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add realZillionX/InspireSkill --skill inspire -a cursorProject install goes to .agents/skills/; add -g for ~/.cursor/skills/.
$ gh skill install realZillionX/InspireSkill inspire --agent cursorProject scope by default (.agents/skills/); add --scope user for a personal install.
Cursor skills documentation · loads skills from .cursor/skills/, .agents/skills/, .claude/skills/, .codex/skills/
Install the "inspire" agent skill from https://github.com/realZillionX/InspireSkill/tree/main into .cursor/skills/inspire/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "inspire", then confirm the skill loads.Cursor copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add realZillionX/InspireSkill --skill inspire -a gemini-cliProject install goes to .agents/skills/; add -g for ~/.gemini/skills/.
$ gh skill install realZillionX/InspireSkill inspire --agent gemini-cliProject scope by default (.agents/skills/); add --scope user for a personal install.
Gemini CLI skills documentation · loads skills from .gemini/skills/, .agents/skills/
Install the "inspire" agent skill from https://github.com/realZillionX/InspireSkill/tree/main into .gemini/skills/inspire/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "inspire", then confirm the skill loads.Gemini CLI copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gh skill install realZillionX/InspireSkill inspireInstalls for Copilot at project scope by default; add --scope user for a personal install. Preview a skill first with gh skill preview. Needs GitHub CLI 2.90.0 or later (public preview).
$ npx skills add realZillionX/InspireSkill --skill inspire -a github-copilotProject install goes to .agents/skills/; add -g for ~/.copilot/skills/.
GitHub Copilot skills documentation · loads skills from .github/skills/, .claude/skills/, .agents/skills/
Install the "inspire" agent skill from https://github.com/realZillionX/InspireSkill/tree/main into .github/skills/inspire/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "inspire", then confirm the skill loads.GitHub Copilot copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add realZillionX/InspireSkill --skill inspire -a opencodeOpenCode documents no install command of its own. Project install goes to .agents/skills/; add -g for ~/.config/opencode/skills/.
$ gh skill install realZillionX/InspireSkill inspire --agent opencodeProject scope by default (.agents/skills/); add --scope user for a personal install.
OpenCode skills documentation · loads skills from .opencode/skills/, .claude/skills/, .agents/skills/
Install the "inspire" agent skill from https://github.com/realZillionX/InspireSkill/tree/main into .opencode/skills/inspire/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "inspire", then confirm the skill loads.OpenCode copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
inspireOperates the Inspire ML platform through its local `inspire` CLI: picking account, workspace and resources, launching notebooks, jobs and services, then cleaning up.
Everything goes through the `inspire` command-line tool of the Inspire (启智) platform. The skill tells the agent to split each task into four parts: scheduling conditions (workspace, project, compute group, quota, resources and image), remote file paths on the shared disk, the workload type, and observation and cleanup. Workloads map to Notebook for interactive debugging, Job for fixed-GPU background runs, HPC for CPU Slurm batches, Ray for elastic or streaming work and Serving for model HTTP services.
Scheduling fields are passed explicitly whenever a workload is created, copied from the live quota and image listings, because the CLI keeps no hidden defaults and treats local caches only as a speed-up. Events, logs, metrics, instances, status and TensorBoard are used to watch a run before results are checked and resources released. Commands, flags and defaults come from `inspire --help`, and focused reference files such as `references/compute-workloads.md` hold the details. The skill text is mostly Chinese.
9 steps, taken from the first numbered list in SKILL.md.
Read from SKILL.md and the folder at commit 5ac8215. It shows what the files ask for, not the result of running them.
Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.
From allowed-tools in the SKILL.md frontmatter.
Ships 1 file in scripts/, which the agent can run.
From the folder's file list and the shell code blocks in SKILL.md.
Links to these hosts (documentation or services it may open):
github.comFrom URLs in SKILL.md, links to its own repository left out.
Names these keys or tokens, usually read from environment variables:
INF_API_KEYFrom names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.
Inspire ML Platform CLI loads about 1.4k tokens when it runs, and up to ~86k if it reads all its reference files. Until then it costs about 84 tokens; SKILL.md has 381 words of instructions outside code blocks.
Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.
The automated check found no risky patterns in SKILL.md.
Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); the scripts in this folder are not scanned.
The full file from realZillionX/InspireSkill at commit 5ac8215, republished under its MIT licence (© realZillionX). 381 words, ~1,392 tokens.
.claude/skills/inspire/SKILL.md (or your agent's skills folder). This skill also uses 571 other files; get the full folder from GitHub.inspire 是启智平台的本地 CLI。命令组、子命令、参数、默认值和示例以 inspire --help、inspire <group> --help 和 inspire <group> <subcommand> --help 为准;本文件只保留平台判断、项目上下文约定、最短执行闭环和 Reference 路由。
inspire account use <name> 设置持久默认账号;所有命令的 --account <name> 仅覆盖本次命令。账号使用本地 Alias。命令开始后固定账号,不受其他进程切换默认账号影响;不传参数只使用默认账号。切账号保留各账号的 Session、SSH Connection 和资源缓存,需要清理时使用相应的显式缓存管理命令。
执行前先把任务拆成四个平面:
| 平面 | 绑定内容 |
|---|---|
| 调度条件 | workspace、project、group、quota、GPU / CPU / 内存 / Shared Memory 和 image,决定任务在哪里、以什么规格运行。 |
| 远端文件 | 代码、数据、权重、Checkpoint 和产物的共享盘绝对路径。可在 shell 里用环境变量缩写;CLI 不维护路径 Alias。数据广场的官方数据集由创建时的 --dataset 只读挂载。 |
| 工作负载 | 交互调试用 Notebook;固定 GPU 后台任务用 Job;CPU Slurm 批处理用 HPC;弹性 Worker、常驻或流式任务用 Ray;模型 HTTP 服务用 Serving。 |
| 观察收尾 | Events 看调度,Logs 看程序(Job / HPC / Ray / Serving 都有,Notebook 用 exec / shell 读),Metrics / Instances 看实际工作单元,Status 看平台状态和落点;读完还要进实例里看的时候 shell 在 Job / HPC / Ray / Serving 下都有,默认进哪个实例按 Workload 定;训练曲线本身用 TensorBoard(inspire tensorboard)读。最后核验业务健康和产物,再清理资源。 |
创建 Workload 时每次显式传入 workspace、project、group、quota 和 image;这些调度字段没有隐式默认值,CLI 也不保存 Workload Profile。选择资源时从同一条 Live Quota Row 复制完整 group 和 quota,并用 Live 镜像目录选当前需要的 image。GPU Job Shared Memory 不能超过所选 Quota 的实例内存;细节见 references/compute-workloads.md。
Live 查询是账号、Workspace、Project、Compute Group、Quota、Image 和资源可用性的事实源;本地缓存只是加速层,可用 inspire cache status|refresh|clear 管理,三条命令都接受 --resource <kind> 只针对一类,不能当作资源事实;refresh 不接受裸形式,必须用 --resource / --workspace / --name 说明刷哪一块,而且正常情况下不需要跑它。CLI 对 Agent 的稳定资源身份只有 Name 和 Alias;同名对象用 Workspace、可读候选和 --pick 消歧。
通用文档只记录平台合同、当前 CLI 行为和长期有效的判断规则。某个账号可见的专属 Workspace / Registry 名称、当次目录数量、即时配额与价格、资源 ID、临时探针结果和待补验结论都不进入通用文档;这些事实随账号和时间变化,使用时从 Live 命令读取。受控验证只保留能够复现且直接约束实现的结论,不保留实验现场或过程统计。
受限 H100/H200 Notebook 的 JupyterTerminal exec 不支持本地 stdin,远端命令的默认 stdin 是 /dev/null。Agent 不得对它使用 --stdin / --bash-stdin,也不得从本地用管道或 < file 向 inspire notebook exec 喂输入;尚未确认 Transport 时同样不要先试。CLI 会在 Transport 预检后、发送用户命令前拒绝这些组合。需要脚本或输入文件时,先把它放到 /inspire/...,再让远端命令用显式重定向或路径读取;需要人工交互时使用带 TTY 的 inspire notebook shell,不要把 shell 当作脚本输入通道。SSH Notebook 的 exec --stdin 不受此限制。
发现类列表和 Batch 结果默认最多展示 20 项,用命令自身的 --limit/-n 收窄或 --all 显式展开;Job 日志默认有行数和字符预算,截断时会给出已展示数量和继续获取完整结果的选项。需要结构化输出时使用根级 inspire --json ...,输出始终是单一 JSON 文档。
Workspace 判断:
CPU资源空间 和 分布式训练空间 是所有用户默认可用的公共 Workspace:前者承担 CPU Notebook、联网准备、依赖安装、HPC 数据处理和 CPU Ray;后者承担 GPU Notebook、GPU Job、多节点训练、Serving 和 GPU 观察。分布式训练空间 通常没有公网:公网内容先在 CPU资源空间 准备,写入共享盘或固化成镜像,再供 GPU Workload 使用;判断细节见 references/internal-sources.md。CLI 不读写仓库级 ./.inspire/,不把仓库绑定到某个 Project,也不假设一个仓库只在一个 Project 中运行。每次操作根据当前任务显式选择 Project / Workspace / 路径 / Image。
INSPIRE.md 仍可作为人类可读的可选资产合同:只在仓库存在需要跨 Agent、成员和会话复用的稳定远端路径、镜像、模型、数据版本或永久基础设施时维护;每项资产自带明确 Project / Workspace 适用范围,可同时覆盖多个 Project。账号凭据、实时状态、日志和短期计划不进入该文件。见 references/assets.md。
INSPIRE.md,先读取其中与当前 Project / Workspace 相关的稳定资产合同;不存在时直接从 Live 事实开始。dry-run 或运行短 Probe。INSPIRE.md。stop,再 delete;终态且不再需要的对象直接 delete,仍有当前或声明的未来消费者时保留。命令语法和参数始终回到 CLI Help。每次先加载最匹配的一份 Reference:
| 用户问题或判断点 | 先加载 |
|---|---|
| 安装、更新、卸载、账号初始化、Windows 原生环境或本机代理设置 | README 快速上手;命令参数以 CLI Help 为准 |
INSPIRE.md、稳定资产身份与生命周期 | references/assets.md |
| Workspace、Compute Group、Quota、实时资源和优先级 | references/resources.md |
项目归属、负责人、预算与平台优先级(inspire project,全局对象、不按 Workspace 划分) | references/assets.md |
| 共享盘、存储池、挂载隔离和绝对路径 | references/paths.md |
| 数据广场检索、官方数据集挂载、版本与访问权限 | references/dataset.md |
| 联网准备、SII 内部源、依赖安装、镜像固化 | references/internal-sources.md |
Notebook 创建、按账号连接、exec / shell / scp、IDE URL、文件流转 | references/notebook.md |
| GPU Job、CPU HPC、Ray、Serving、TensorBoard,及提交后的观察、优先级与异常 | references/compute-workloads.md |
| CPU 准备、数据处理、GPU 训练、部署或交付的阶段化计划 | references/workflows.md |
| Image 选择、保存、注册、可见性和清理 | references/image.md |
Serving Endpoint、INF_API_KEY、API Key 管理与请求亲和性 | references/compute-workloads.md |
| Model Registry、Model Version 与 Serving 的关系 | references/model.md |
仅在维护 CLI Browser API 封装、核对前端请求合同,或用户明确要求接口细节时加载:
| 需要什么 | 加载 |
|---|---|
请求契约、响应信封、认证与 Session、分页、Workspace scoping、错误码、探针方法、变更验收;当前 CLI 使用的 Action 请求体 / 响应 / 参数语义 / CLI 映射 / 限制;创建面字段合同;数据广场(aip.sii.edu.cn)的握手、信封与目录端点 | references/dev/browser-api.md |
© realZillionX, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file
SKILL.md and 571 other files (scripts, references, assets) in the repository root of realZillionX/InspireSkill.
Open the folder on GitHubat commit 5ac8215
Inspire ML Platform CLI next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.
| Skill | Stars | Used in | Tokens | Auto-check | Licence | Repo updated |
|---|---|---|---|---|---|---|
| Inspire ML Platform CLI this skillrealZillionX/InspireSkill | 551 | — | ~1.4k | Automated safety check: Pass | MIT | |
| SkyPilot Multi-Cloud OrchestrationOrchestra-Research/AI-Research-SKILLs | 13k | 4 repos | ~2.4k | Automated safety check: Pass | MIT | |
| Megatron-LM on SLURMNVIDIA/Megatron-LM | 18k | — | ~1.8k | Automated safety check: Pass | Apache-2.0 | |
| Lambda Labs GPU CloudOrchestra-Research/AI-Research-SKILLs | 13k | 4 repos | ~3k | Automated safety check: Warn | MIT | |
| Lambda LabsLuciole-Studio/Misaka-Agent | 171 | 1 repos | ~3k | Automated safety check: Warn | MIT | |
| Weights & Biases Experiment TrackingOrchestra-Research/AI-Research-SKILLs | 13k | 9 repos | ~3.1k | Automated safety check: Pass | MIT |
Orchestra-Research/AI-Research-SKILLs
Runs ML training and batch jobs across clouds with SkyPilot, using spot instances, automatic region selection and managed recovery to cut GPU cost.
NVIDIA/Megatron-LM
Shows how to launch distributed Megatron-LM training on a SLURM cluster: sbatch skeleton, torch.distributed.run setup, CUDA_DEVICE_MAX_CONNECTIONS rules and failure diagnosis.
Orchestra-Research/AI-Research-SKILLs
Guide to renting GPUs on Lambda Labs for ML training and inference: on-demand instances, 1-Click Clusters, SSH access, persistent filesystems and alternatives.
Luciole-Studio/Misaka-Agent
On-demand GPU cloud instances for ML training. An agent skill from Luciole-Studio/Misaka-Agent.
Orchestra-Research/AI-Research-SKILLs
Guides an agent through tracking ML experiments with W&B: run logging, config capture, hyperparameter sweeps, artifacts and a model registry.
Luciole-Studio/Misaka-Agent
Serverless GPU cloud for ML jobs and model APIs. An agent skill from Luciole-Studio/Misaka-Agent.
Categories
Operates the Inspire ML platform through its local `inspire` CLI: picking account, workspace and resources, launching notebooks, jobs and services, then cleaning up. Everything goes through the `inspire` command-line tool of the Inspire (启智) platform. The skill tells the agent to split each task into four parts: scheduling conditions (workspace, project, compute group, quota, resources and image), remote file paths on the shared disk, the workload type, and observation and cleanup.
Inspire ML Platform CLI fits situations like: starting a GPU job or notebook on the Inspire platform; choosing a workspace, project, quota and image for a new workload; reading logs, metrics and status of a running job, then cleaning up; deploying a model as an HTTP service on Inspire.
Run `npx skills add realZillionX/InspireSkill --skill inspire -a claude-code`. Or copy the skill folder (the realZillionX/InspireSkill repository) into .claude/skills/inspire in your project. Claude Code loads it when a task matches its description.
Run `npx skills add realZillionX/InspireSkill --skill inspire -a codex`. Or copy the skill folder (the realZillionX/InspireSkill repository) into .agents/skills/inspire in your project. Codex loads it when a task matches its description.
Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add realZillionX/InspireSkill --skill inspire -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/inspire, .gemini/skills/inspire, .github/skills/inspire and .opencode/skills/inspire in your project.
Going by SKILL.md and its folder, Inspire ML Platform CLI needs credentials named INF_API_KEY. Our summary lists: The `inspire` command-line tool; An account on the Inspire platform.
SKILL.md names 1 domain. As links in the text: github.com. This is read from the text; nothing was executed.
Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. The check reads SKILL.md only: the scripts in the folder are not scanned, so read them before running anything.
Inspire ML Platform CLI is published under the MIT licence (from the LICENSE file in the skill folder). It allows redistribution, so the full SKILL.md is shown on this page.
About 1.4k tokens (SKILL.md is roughly 5.6k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full. Its references folder adds about 85k tokens, read only when the agent opens those files.
Skills that share tags, products or a category with Inspire ML Platform CLI: SkyPilot Multi-Cloud Orchestration (Orchestra-Research/AI-Research-SKILLs, 13k stars), Megatron-LM on SLURM (NVIDIA/Megatron-LM, 18k stars), Lambda Labs GPU Cloud (Orchestra-Research/AI-Research-SKILLs, 13k stars) and Lambda Labs (Luciole-Studio/Misaka-Agent, 171 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.
realZillionX (a GitHub user) maintains it in realZillionX/InspireSkill, which has 551 GitHub stars. The repository was last updated on October 8, 2026.
Source: realZillionX/InspireSkill on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.