sealos-skills
Agent plugin for deploying and operating apps on Sealos Cloud,
packaged in the Agent Plugins format and as a
DeepSeek Harness profile bundle.
The same use-sealos skill is also installable through Codex, Claude Code,
Cursor, Qoder, skills.sh, OpenClaw / ClawHub, CodeBuddy, Gemini CLI, Qwen
Code, and generic repo importers.
Say "deploy X to Sealos" (or "帮我把 X 部署到 Sealos") in a compatible agent
and the use-sealos skill handles the rest: sign-in, picking the fastest
deploy path, databases, storage, public HTTPS, and post-deploy verification.
What's inside
package.json # DeepSeek Harness bundle manifest (`dsh.bundle`)
cordis.patch.yml # inserts the skill provider into a dsh profile
index.js # registers `use-sealos` on `ctx.skills`
CLAUDE.md # context-only hosts (Gemini, Qwen, Amp / Kimi)
gemini-extension.json # Gemini CLI extension
qwen-extension.json # Qwen Code extension
openclaw.plugin.json # unused on OpenClaw's native package.json path; install plugins/sealos instead
.codex-plugin/plugin.json # root Codex import manifest
.agents/plugins/marketplace.json # Codex marketplace entry
.codebuddy-plugin/marketplace.json # CodeBuddy marketplace
skills/ # skills.sh entries (symlinks → plugins/sealos/skills/*)
plugins/sealos/
├── .codex-plugin/plugin.json # Codex marketplace plugin manifest
├── .claude-plugin/plugin.json # Claude Code manifest
├── .cursor-plugin/plugin.json # Cursor manifest
├── .qoder-plugin/plugin.json # Qoder manifest
├── commands/sealos.md # /sealos for compatible plugin hosts
└── skills/
├── use-sealos/
│ ├── SKILL.md # intent router + execution rules
│ ├── references/ # loaded on demand
│ │ ├── deploy.md # three deploy paths (store / image / source)
│ │ ├── build.md # Dockerfile, linux/amd64 build, registry push
│ │ ├── databases.md # KubeBlocks clusters + credentials
│ │ ├── platform.md # manifest contract (labels, Ingress, storage, quota)
│ │ ├── operate.md # status, logs, debugging, deletion
│ │ └── recipes.md # per-app recipes for popular self-hosted software
│ └── scripts/
│ ├── sealos-api.py # auth (OAuth2 device flow) + Template API, stdlib only
│ └── wait-app.sh # post-deploy readiness + URL verification
├── sealos-deploy/SKILL.md # Brain managed-mode entry (MCP-gated pipeline)
└── k8s-kaniko-job/ # sandbox build executor (Kaniko Job → GHCR)
├── SKILL.md
└── scripts/kaniko-build.py
Installation
Cursor
From a published GitHub repo: Settings → Plugins, paste the repository
URL in Search or Paste Link, open the Sealos plugin, click Add to
Cursor.
For local development, symlink the skill into your personal skills folder:
ln -sfn "$(pwd)/plugins/sealos/skills/use-sealos" ~/.cursor/skills/use-sealos
Claude Code
claude plugin marketplace add labring/sealos-skills
claude plugin install sealos@sealos-skills
/reload-plugins
Use /sealos for interactive work. The marketplace also accepts a local
checkout path.
Codex
codex plugin marketplace add labring/sealos-skills
codex plugin add sealos@sealos
In Codex CLI, invoke $sealos; in Codex App, select Sealos under Plugins.
The marketplace source is .agents/plugins/marketplace.json; the plugin
manifest is plugins/sealos/.codex-plugin/plugin.json. A repository-root
Codex manifest is also included for direct importers.
Qoder
In Qoder, open the plugin Marketplace (Settings → Plugins → Marketplace),
click + Create Plugin, choose import from a local folder, and select
plugins/sealos from a clone of this repository.
To distribute, zip the contents of plugins/sealos (the zip root must
contain .qoder-plugin/plugin.json) as sealos-2.0.0.zip and share or
publish it through the Qoder marketplace.
Use /sealos in Qoder after installation.
DeepSeek Harness
This repository is a dsh profile bundle. After npx @deepseek-ai/dsh web works, install the plugin 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
use-sealos appears in the session skill catalog. Say "deploy X to Sealos"; the model loads the skill via the skill tool, then runs scripts/sealos-api.py / kubectl through bash.
Login writes ~/.sealos/kubeconfig; grant the host's required filesystem
access when prompted.
Add the GitHub topic dsh-plugin on the public repo so it shows up in the harness plugin index.
OpenClaw
Install the plugin directory, not the repository root:
openclaw plugins install /path/to/sealos-skills/plugins/sealos
OpenClaw loads it as a Cursor bundle and exposes use-sealos. Slash-command
exposure is not claimed. Say "deploy X to Sealos".
Do not install the repo root. OpenClaw sees root package.json first, treats
the checkout as a native plugin, and fails because that package has no
openclaw.extensions. Root openclaw.plugin.json is not used on that path.
CodeBuddy
/plugin marketplace add labring/sealos-skills
Command exposure depends on the CodeBuddy runtime.
Gemini CLI
Context-only extension. Slash commands are not claimed.
gemini extensions install https://github.com/labring/sealos-skills
Ask Gemini to deploy to Sealos Cloud. It loads CLAUDE.md, which points at
use-sealos.
Qwen Code
Context-only extension. Slash commands are not claimed.
qwen extensions install https://github.com/labring/sealos-skills
skills.sh
npx skills add labring/sealos-skills
The pack installs three skills: use-sealos (the interactive entry — say
"deploy X to Sealos"), plus sealos-deploy and k8s-kaniko-job, which are
inert outside a managed sandbox (see below).
Brain (managed sandbox)
Brain's GitHub-deploy pipeline installs this pack into a Devbox with
npx skills add <repo> -y and drives the sealos-deploy skill in managed
mode (SEALAI_DEPLOY_MODE=managed, MCP handshake via template_ready /
deployment_completed, source builds via a k8s-kaniko-job Kaniko Job).
Point Brain's DEPLOY_SKILL_SOURCE env at this repository. The design and
the full control-plane contract live in BRAIN-ADAPTATION.md.
Amp / Kimi / generic repo importers
Import https://github.com/labring/sealos-skills.git. Behavior is
host-dependent. Context is CLAUDE.md; the skill lives at skills/use-sealos
and plugins/sealos/skills/use-sealos.
First run
Say "deploy X to Sealos". The skill checks credentials
(sealos-api.py status) and, if needed, signs you in via OAuth2 device flow —
or paste a kubeconfig from the Sealos web console into ~/.sealos/kubeconfig.
Deploy strategy
- Template store — 200+ one-click templates, deployed via the Template
API (~3 min).
- Official Docker image — a generated template (workloads + KubeBlocks
databases + Ingress) for self-hosted apps not in the store (~5-10 min).
- Source build — Dockerfile →
docker buildx --platform linux/amd64 →
registry push → image deploy (for the user's own projects).
All three paths go through the Sealos Template API, so every deployment is
tracked as an instance: visible in the Sealos UI and removable as a unit.
Requirements
kubectl, python3 (stdlib only); docker + gh only for the source path- A Sealos account — the skill signs in via OAuth2 device flow, or paste a
kubeconfig from the Sealos web console into
~/.sealos/kubeconfig
Development
Scratch clones, test dumps, and other local-only artifacts go in
validation-assets/ (gitignored).
This 2.0 release replaces the previous eight-skill implementation. The
standalone Canvas, Sealos Desktop app builder, cloud-native readiness,
Dockerfile, and Compose-conversion skills are archived on the
codex/backup-before-next-20260928 branch. Deployment, database, object
storage, and operations now route through use-sealos; Brain's managed
deployment has its own sealos-deploy and k8s-kaniko-job skills.
