为 DeepSeek Harness 打包 8 个 Sealos Cloud 技能,覆盖部署、云数据库、对象存储与本地只读拓扑视图。
$ dsh plugin --profile web add github:labring/sealos-skills在终端中运行以上命令,通过 dsh CLI 安装此插件。可在右上角切换 Profile。 第一次用 dsh?看这篇新手教程
把 Sealos Cloud 团队维护的 8 个 Agent Skill 一次性打包进 DeepSeek Harness 的 web profile,让 AI agent 在会话里直接完成项目部署、云数据库、S3 对象存储、Sealos Desktop 应用开发等操作。
@deepseek-ai/cordis (peer, 由宿主注入)、yaml ^2.4.2 (用于解析 SKILL.md 的 YAML frontmatter)inject = ['skills'] 依赖宿主 ctx.skills,调用 ctx.skills.registerProvider() 注册名为 sealos 的 provider,rank=600(与宿主内置 provider 同级)。cordis.patch.yml 在 bundle 阶段追加 sealos-skills 节点,依赖 @deepseek-ai/dsh-base 已挂载的技能注册中心。正在用 DeepSeek Harness 想让 AI agent 直接处理 Sealos Cloud 任务的人。如果只想让 agent 写 Dockerfile、读懂 docker-compose.yml、或者已经部署过项目想看本地资源拓扑图,这个 bundle 也单独可用,但最有价值的还是把"代码→部署→看效果"这条链全跑通。
| 依赖 | 最低版本 | 说明 |
|---|---|---|
| DeepSeek Harness | 未声明 | README 文档示例为 npx @deepseek-ai/dsh web,需已安装并初始化 web profile |
| Node.js | 未声明 | index.js 使用 node:fs/promises 与顶层 await;test.js 使用 node:test,建议 Node 18+ |
| pnpm | 未声明 | README 第103行说明 bundle 安装时 pnpm 必须位于 PATH |
| 平台 | 跨平台 | 纯 JS 实现,未声明操作系统限制 |
| 原生模块 | 无 | 仅依赖 yaml,无 node-gyp / 原生扩展 |
| 运行时外部工具 | 按需 | sealos-deploy metadata.compatibility 要求:Sealos 账号/workspace、docker、buildx、gh CLI(按需)、git(按需)、kubectl、Node 18+(可选加速)、Python 3.8+ 与 PyYAML(Phase 5)、kompose/crane(Compose 转换时) |
dsh plugin --profile web add github:labring/sealos-skills
本插件无需额外配置。bundle 在启动时自动扫描仓库内 skills/*/SKILL.md,注册全部 8 个技能到 ctx.skills。
Q: 安装后怎么在 DSH 里调用?
A: DSH 没有为这个 bundle 提供 /sealos 斜杠命令,直接告诉 agent "部署到 Sealos Cloud"、"创建云数据库"、"接 S3" 或要求查看上次部署的资源,agent 会通过 skill tool 加载对应的 sealos-* 技能并执行。
Q: 必须安装到 web profile 吗?
A: 是的。README 第103行明确说明这个仓库是 web profile 的 dsh bundle,安装到其他 profile 不会生效。
Q: 使用前要准备哪些外部工具?
A: 真要部署项目到云端,需要 docker、buildx、kubectl、gh CLI(按需)、sealos-cli;纯生成 Dockerfile 或把 docker-compose 转 Sealos 模板只需要 Node.js 和 Python 3.8+。sealos-deploy 的 metadata.compatibility 字段列出了完整清单。
Q: 需要 Sealos Cloud 账号才能用吗?
A: 不一定。云数据库、对象存储、实际部署都需要 Sealos 账号与 workspace;只生成 Dockerfile、转 Compose 模板、查看 canvas 都不需要云端账号。
Q: DSH 默认 bash 沙箱够用吗?
A: 不够。登录 Sealos 会写 ~/.sealos/kubeconfig,默认沙箱禁止该路径,需要在 DSH 配置中启用 sandbox_permissions: danger-full-access,否则登录步骤会被沙箱拦截。
Q: 怎么升级或卸载?
A: 升级:再次执行 dsh plugin --profile web add github:labring/sealos-skills 会覆盖式更新 skills/。卸载:在 profile 中移除这个 bundle 或直接删除 skills/ 目录,DSH 没有为它提供专用的 uninstall 命令。
Q: 8 个技能是怎么决定优先级的?
A: 全部使用 BUNDLED_SKILL_RANK=600(index.js 第8行),与 DSH 内置技能 provider 同级,按技能名升序排列。
入门 — bundle 本身无需配置,安装即可在会话中通过自然语言触发;但真正完成云端部署需要 Sealos 账号、容器仓库、sealos-cli 与 kubectl,第一次完整跑通前会有较多外部准备工作。
English | 简体中文 | 繁體中文 | 日本語 | 한국어 | Español | Français | Deutsch | Português (Brasil) | Русский | العربية | हिन्दी | Bahasa Indonesia
Deploy projects to Sealos Cloud from your AI agent.
Sealos Skills is a plugin-first skill pack centered on Sealos Cloud development and deployment. It helps an AI agent inspect a project, prepare missing deployment artifacts, connect Sealos Cloud databases and object storage for development, build or reuse a container image, ship the app to Sealos Cloud, and view deployed resources in a local read-only canvas.
The recommended Codex path is native Codex plugin installation. Cross-host plugin installs, skills.sh, and context-only extension hosts such as Gemini CLI and Qwen Code use the same root skills/** source.
Add this repository as a Codex marketplace, then install the Sealos plugin:
codex plugin marketplace add labring/sealos-skills
codex plugin add sealos@sealos
One Sealos plugin installs the deploy, database, S3, canvas, app-builder, and supporting cloud-native skills from root skills/**: sealos-deploy, sealos-database, sealos-s3, sealos-canvas, sealos-app-builder, cloud-native-readiness, dockerfile-skill, and docker-to-sealos.
For compatibility and local Codex testing, install the same plugin with:
npx plugins add https://github.com/labring/sealos-skills --target codex
After installation in Codex, use the plugin from Codex:
$sealos
Codex examples:
$sealos deploy this repo to Sealos Cloud
$sealos deploy /path/to/project
$sealos deploy https://github.com/labring-sigs/kite
$sealos create a cloud Postgres database for this repo and wire DATABASE_URL
$sealos create private S3 object storage for uploads and wire env vars
Add this repository as a Claude Code marketplace, then install the Sealos plugin:
claude plugin marketplace add labring/sealos-skills
claude plugin install sealos@sealos
For compatibility with cross-host plugin installers, install the same plugin with:
npx plugins add https://github.com/labring/sealos-skills --target claude-code
If you only use one detected agent tool on the machine, you can let plugins choose the target:
npx plugins add https://github.com/labring/sealos-skills
After installation in Claude Code, use /sealos:
/sealos deploy this repo to Sealos Cloud
/sealos deploy /path/to/project
/sealos deploy https://github.com/labring-sigs/kite
/sealos create a cloud Postgres database for this repo and wire DATABASE_URL
/sealos create private S3 object storage for uploads and wire env vars
Build the Qoder plugin package from the repository root:
python3 scripts/package-qoder-plugin.py
Import dist/sealos-1.2.5.zip into Qoder. The package exposes the same eight root-level skills as the Codex plugin and provides /sealos as its command entry point.
Qoder examples:
/sealos deploy this repo to Sealos Cloud
/sealos create a cloud Postgres database for this repo and wire DATABASE_URL
/sealos create private S3 object storage for uploads and wire env vars
/sealos show the resources created by the last deployment
This repository is a dsh profile bundle over the same root skills/** source. After npx @deepseek-ai/dsh web works, install it into the same web profile (pnpm must be on PATH):
npx @deepseek-ai/dsh plugin --profile web add github:labring/sealos-skills
npx @deepseek-ai/dsh web
A local checkout:
npx @deepseek-ai/dsh plugin --profile web add /path/to/sealos-skills
The eight root skills appear in the session skill catalog. Ask the agent to deploy to Sealos Cloud; it loads sealos-deploy (and sibling skills as needed) through the skill tool, then runs kubectl / sealos-cli through bash.
The default bash sandbox blocks writes outside the workspace. Login writes ~/.sealos/kubeconfig, so those commands need sandbox_permissions: danger-full-access.
| Tool | Install | Usage |
|---|---|---|
| Codex CLI / Codex App | codex plugin marketplace add labring/sealos-skills then codex plugin add sealos@sealos | $sealos in Codex CLI, or + → Plugins → Sealos in Codex App |
| Claude Code | claude plugin marketplace add labring/sealos-skills then claude plugin install sealos@sealos | /sealos |
| Claude Code compatibility path | npx plugins add https://github.com/labring/sealos-skills --target claude-code | /sealos |
| Qoder | Build with python3 scripts/package-qoder-plugin.py, then import the ZIP | /sealos or automatic skill selection |
| DeepSeek Harness | npx @deepseek-ai/dsh plugin --profile web add github:labring/sealos-skills | Ask the agent to use Sealos skills; there is no /sealos slash command |
| OpenClaw / ClawHub | clawhub install labring/sealos-skills | Host command exposure depends on the ClawHub runtime |
| CodeBuddy | /plugin marketplace add labring/sealos-skills | Host command exposure depends on the CodeBuddy runtime |
| Gemini CLI | gemini extensions install https://github.com/labring/sealos-skills | Context-only extension; ask Gemini to use Sealos Skills |
| Qwen Code | qwen extensions install https://github.com/labring/sealos-skills | Context-only extension; ask Qwen to use Sealos Skills |
| Amp / Kimi / generic repo importers | Import https://github.com/labring/sealos-skills.git | Host-dependent |
Gemini CLI and Qwen Code manifests provide repository context through CLAUDE.md; they do not claim slash-command support.
skills.sh skill packIf your agent uses skills.sh directly, install the same skills pack with:
npx skills add labring/sealos-skills
Then run the deploy skill directly:
/sealos-deploy
/sealos-deploy /path/to/project
/sealos-deploy https://github.com/labring-sigs/kite
/sealos-database create a cloud Postgres database for this repo and wire DATABASE_URL
/sealos-s3 create private object storage for uploads and wire env vars
After a project has been deployed and .sealos/state.json contains verified last_deploy runtime evidence, use the sealos-canvas skill through your installed plugin entry point for a local read-only view. Canvas stays plugin-pack mediated because it consumes verified deployment state.
/sealos-deploy, /sealos-database, and /sealos-s3 are direct skills.sh skill entries. Plugin usage should go through $sealos in Codex or /sealos in Claude Code.
Prefer the plugin install for Codex, Claude Code, and Qoder because it:
The Codex integration follows OpenAI's Codex plugin build guide:
.codex-plugin/plugin.json contains plugin identity, discovery metadata, interface copy, default prompts, brand metadata, and asset paths relative to the repository root..agents/plugins/marketplace.json registers this repo-local plugin for local Codex marketplace testing..claude-plugin/plugin.json and .claude-plugin/marketplace.json define the Claude Code-compatible plugin surface..qoder-plugin/plugin.json defines the Qoder plugin surface and explicitly exposes all eight Codex skills.qoder.md provides Qoder-level routing and safety instructions without copying skill implementations.distribution/platforms.json records platform support claims and evidence.marketplaces/README.md owns marketplace rules and prevents command-support overclaims.scripts/validate-codex-plugin.py validates the Codex manifest, Claude Code metadata, repo marketplaces, platform registry, and asset paths.scripts/package-qoder-plugin.py builds a Qoder-compatible ZIP with the plugin manifest at the archive root.skills/**/SKILL.md remains the only skill source; do not add a second packaged copy of the skills.package.json, cordis.patch.yml, and index.js register that same root skill tree as a DeepSeek Harness profile bundle.Validate plugin metadata before publishing or pushing manifest changes:
python3 scripts/validate-codex-plugin.py
python3 -m json.tool .codex-plugin/plugin.json >/dev/null
python3 -m json.tool plugin.json >/dev/null
python3 -m json.tool .agents/plugins/marketplace.json >/dev/null
python3 -m json.tool marketplace.json >/dev/null
python3 -m json.tool .claude-plugin/plugin.json >/dev/null
python3 -m json.tool .claude-plugin/marketplace.json >/dev/null
python3 -m json.tool .qoder-plugin/plugin.json >/dev/null
python3 -m json.tool distribution/platforms.json >/dev/null
You only need a plugin-compatible or skills.sh compatible AI agent and a project to deploy.
During the deploy, database, and object-storage flows, Sealos Skills will:
kubectl are availablesealos-cli for Sealos Cloud database creation, connection details, and database operationssealos-cli s3 for Sealos object storage buckets, credentials, quota checks, object operations, and presigned URLsFor an actual deployment, you will still need a Sealos Cloud account and access to a container registry, but these do not need to be fully set up before the skill starts. For database and object-storage work, you need a Sealos Cloud account and a workspace that can create the requested resources.
On a typical deploy, the agent will:
Later runs can switch to an in-place update flow when an existing deployment is detected.
For a local project or Devbox that needs a cloud database, the agent will:
DATABASE_URL, Prisma, Drizzle, MongoDB, MySQL, or Redissealos-cli database to list, create, inspect, and connect Sealos Cloud databasesFor a local project or Devbox that needs S3-compatible object storage, the agent will:
sealos-cli s3 from zjy365/sealos-cli#28 to list, create, inspect, and update object storage bucketsFor a repository already deployed by Sealos Deploy, the agent will:
.sealos/state.json to locate the deployed app.kubectl get commands.127.0.0.1 canvas UI.If the project has not been deployed yet, Sealos Canvas stops and directs the user to deploy the project first.
The plugin and skills.sh pack expose the same skill source:
sealos-deploy — deploy a local or GitHub project to Sealos Cloudsealos-database — create, connect, and operate Sealos Cloud databases for developmentsealos-s3 — create buckets, connect credentials, check quota, and operate Sealos S3-compatible object storagesealos-canvas — view deployed Sealos resources in a local read-only canvas UIsealos-app-builder — build Sealos Desktop apps with SDK integrationcloud-native-readiness — assess deployment readinessdockerfile-skill — generate production-ready Dockerfilesdocker-to-sealos — convert Docker Compose services into Sealos templatesskills/ is the single source of truth for Sealos deploy, Sealos canvas, and the supporting skills used during the deploy flow. The same root-level skills directory serves skills.sh installs and every plugin or extension manifest in this repository.
Important distribution files:
.codex-plugin/plugin.json — Codex plugin manifest.agents/plugins/marketplace.json — local Codex marketplace entry.claude-plugin/plugin.json — Claude Code-compatible plugin manifest.qoder-plugin/plugin.json — Qoder plugin manifestqoder.md — Qoder plugin routing and safety instructionsmarketplace.json and .claude-plugin/marketplace.json — Claude-compatible marketplace entries.codebuddy-plugin/marketplace.json — CodeBuddy marketplace entrygemini-extension.json — Gemini CLI context extensionqwen-extension.json — Qwen Code context extensionopenclaw.plugin.json — OpenClaw / ClawHub bundle pointerpackage.json, cordis.patch.yml, and index.js — DeepSeek Harness profile bundle; registers root skills/** on ctx.skillscommands/sealos.md — /sealos plugin command entry for compatible hostsdistribution/platforms.json — platform support registrymarketplaces/README.md — marketplace rules and support-claim ownershipscripts/validate-codex-plugin.py — Codex plugin validationscripts/package-qoder-plugin.py — Qoder ZIP packagerDo not add a second packaged copy of the skills. Root skills/** is the only skill source for all installation paths.
MIT