Preface

Most of the coding work can already be handled by most modern agents today. What often really stalls progress is the latter half of the process: whether there is a Dockerfile, where to push the image, how to write Kubernetes templates, how to provision databases and object storage, and how to connect environment variables to the project. These steps are scattered across multiple CLIs and documents, and it is difficult to complete them all in one session.

The core philosophy of DeepSeek Harness (hereinafter referred to as DSH) is “everything is a plugin”: models, tools, skills, and sandboxes can all be replaced via plugins without modifying the Harness source code. A community member has created an independent site for the plugin index: DeepSeek Harness Plugin Repository, which has no official affiliation with DeepSeek or HyperFine, and should not be treated as an official app store. There is an entry named sealos-skills under the “Development & Runtime” category in the directory, which aims to connect Sealos Cloud’s deployment, database, and object storage capabilities to agents.

What is This

sealos-skills is maintained by the GitHub organization labring, and its repository is labring/sealos-skills. The positioning given on the directory page is: Sealos agent skills, deploy any project with one command, provision databases and object storage, compatible with Claude Code, Codex, etc.

To be more precise, it is not a single-function plugin, but a plugin-first skill pack. The skills/** in the root directory is the sole skill source; the same source will be registered as a Codex/Claude Code plugin, and also registered as a DSH profile bundle via package.json, cordis.patch.yml and index.js, mounting the skills onto ctx.skills. The version number in package.json is 1.2.5, the package name is sealos, and the plugin export name is sealos-skills.

As of 2026-08-17, both the directory page and the GitHub repository have 70 stars. The directory page shows that the main language is Python, the latest push date is 2026-08-14, and the inclusion date is 2026-08-15. The README and package.json state the license as MIT; there is no separate LICENSE file in the repository root, and the GitHub license field is currently empty, so you should verify the license yourself before installing.

Sealos itself is a Kubernetes-powered cloud operating system developed by labring, with its public cloud entry at sealos.io. The tasks the skill pack helps agents complete are: inspect projects, complete missing deployment artifacts, connect to Sealos Cloud databases and object storage, build or reuse container images, publish applications to Sealos Cloud, and view deployed resources in a local read-only canvas.

Core Features

A Sealos plugin will install these eight skills from the skills/** directory:

  1. sealos-deploy: Deploy a local directory or GitHub repository to Sealos Cloud. The skill documentation specifies that supported workload types are server, static-web, worker, scheduled-job, and approved remote-desktop; targets such as desktop, mobile, CLI, libraries, dependencies requiring specific hardware, hybrid, or unrecognized workloads will be rejected before scoring or building. The typical workflow is to evaluate project structure and runtime requirements, reuse existing images or build when needed, generate Sealos templates, perform Runtime Truth verification after deployment: check the actual Sealos App URL, routing, web application login or setup process, logs, and resource scope. If an existing deployment is detected on subsequent runs, it will switch to in-place updates.
  2. sealos-database: Provision a Sealos Cloud database for a local project or Devbox. It will detect signals such as DATABASE_URL, Prisma, Drizzle, MongoDB, MySQL, Redis, etc., then use sealos-cli database to list, create, check, and connect. It only writes environment variable keys already in use by the project, and will not print passwords or full connection strings in chat. The CLI-supported types include postgresql, mongodb, mysql, apecloud-mysql, redis, kafka, qdrant, nebula, weaviate, milvus, pulsar, clickhouse; postgresql is the default when the project has no clear signal. Public network access and deletion operations require prior confirmation.
  3. sealos-s3: Provision S3-compatible object storage. It will detect S3 environment variables, AWS SDK, MinIO, upload paths, or pre-signed URL code, then use sealos-cli s3 to manage buckets, credentials, quotas, and object operations. Private buckets are created by default, and only the minimum necessary environment variables are written. Public buckets or credential rotation require confirmation.
  4. sealos-canvas: For repositories that have already been deployed via Sealos Deploy, read .sealos/state.json, query the namespace with a read-only kubectl get, and start a temporary topology canvas on 127.0.0.1. It will pause and prompt to perform a deployment first if the project has not been deployed yet.
  5. sealos-app-builder: Connect web applications to Sealos Desktop using the Sealos Desktop SDK (@labring/sealos-desktop-sdk), handling SDK initialization, session authentication, and local iframe debugging.
  6. cloud-native-readiness: Perform a read-only assessment of whether a repository is suitable for cloud deployment. First perform eligibility checks, then score 0–12 points (0–2 per category) across six dimensions: stateless, externalized configuration, horizontal scaling, start/stop, observability, and service boundaries. Unsupported workloads will not proceed to the build phase.
  7. dockerfile-skill: Generate production-ready Dockerfiles through analysis → generation → build repair → runtime verification. The skill explicitly states that you cannot only verify that docker build succeeds, but also validate migrations, HTTP/health checks, and logging.
  8. docker-to-sealos: Convert Docker Compose or installation documentation into Sealos templates. This step does not make online changes by itself; pass the validation before handing it off to sealos-deploy.

During deployment, database, and object storage workflows, the skills will check if Docker and kubectl are available, guide Sealos login when needed, and execute operations via sealos-cli / sealos-cli s3. Actual deployment still requires a Sealos Cloud account, as well as access to container image registries such as Docker Hub and GHCR; these can be prepared after the skill starts running. Databases and object storage also require the current workspace to have permissions to create corresponding resources.

Installation and Activation

The installation command given on the directory page is to run this in the DeepSeek Harness terminal:

dsh plugin add github:labring/sealos-skills

For reproducible installations, pin the commit hash as per the directory page instructions:

dsh plugin add github:labring/sealos-skills#commit

Replace #commit with the actual commit hash.

The repository README also includes an installation method for the DSH web profile. The prerequisites are that npx @deepseek-ai/dsh web is already functional, and pnpm is in the PATH:

npx @deepseek-ai/dsh plugin --profile web add github:labring/sealos-skills
npx @deepseek-ai/dsh web

If you have a locally checked out repository, you can replace the GitHub address with the local path:

npx @deepseek-ai/dsh plugin --profile web add /path/to/sealos-skills

After installation, the eight root skills mentioned above will appear in the session’s skill directory. DSH does not have a /sealos slash command, you need to use natural language to instruct the agent to use the Sealos skills; it will load sealos-deploy and required supporting skills via the skill tool, then run kubectl / sealos-cli via bash.

The same repository can also be installed to hosts such as Codex, Claude Code, Qoder, Gemini CLI, Qwen Code, etc., with the skill source still being skills/**. The entry points in those environments are $sealos or /sealos, which is different from DSH, so do not directly copy the slash commands.

Typical Usage

In a DSH session, you can directly say phrases like the following, consistent with the examples in the repository README:

Deploy the current repository to Sealos Cloud
Deploy /path/to/project to Sealos Cloud
Deploy https://github.com/labring-sigs/kite to Sealos Cloud
Create a cloud Postgres database for this repository and write the DATABASE_URL
Create a private S3 object storage for upload functionality and write environment variables
View the resources created by the last deployment

The sample GitHub repository given in the README is labring-sigs/kite.

sealos-deploy requires that you have completed Sealos login and selected a workspace. Docker, buildx, and the gh CLI are only required when performing local builds and pushes; git is only needed when cloning from a GitHub URL or using git metadata. Phase 5 also requires Python 3.8+ and PyYAML; if there is a Compose file in the root directory that needs conversion, kompose may be used, and crane may be needed when the image tag is a floating tag. Node.js 18+ is listed in the documentation as an optional acceleration item, not a hard prerequisite.

The default bash sandbox prohibits writing to paths outside the workspace. Sealos login will write to ~/.sealos/kubeconfig, and the README explicitly states that these commands require sandbox_permissions: danger-full-access.

Applicable Scenarios and Notes

This is suitable for the following scenarios:
1. You are already using DSH for coding and want to deploy deployable web/server projects to Sealos Cloud in the same session.
2. Local or Devbox development requires real cloud databases and S3-compatible storage, rather than relying on Postgres/MinIO in Compose all the time.
3. Your project already has a docker-compose.yml and wants to convert it to a Sealos template before deployment.
4. After deployment, you want to view the resource topology locally instead of directly modifying the cluster.

Precautions before use:
1. The plugin runs with the permissions of the current dsh process, and may execute code during installation. You should inspect the source code repository and license before installing.
2. The community directory is an independent site and not an official DeepSeek app store; sealos-skills is an open-source skill pack from labring.
3. DSH does not have a /sealos command; the usage is to instruct the agent to load the skill and then run the CLI.
4. Login will bypass the default sandbox and write files to ~/.sealos/kubeconfig, requiring explicit danger-full-access permissions.
5. The directory introduction says “deploy any project with one command”, but the skill documentation has stricter eligibility checks: unsupported workloads such as desktop, mobile, pure libraries, CLI, etc., will be rejected.
6. Database passwords, S3 keys, full connection strings, and kubeconfig should not appear in chat replies according to the skill conventions; .env / .env.local should also not be committed to the repository.
7. Public network databases, public buckets, deletion, rollback, and cleanup operations all require prior confirmation.
8. sealos-canvas is read-only and relies on verified last_deploy evidence in .sealos/state.json; do not expect it to draw a topology out of thin air without a prior deployment.

Summary

sealos-skills bundles Sealos Cloud’s deployment, database, object storage, and local read-only canvas into a single skill pack, connecting it to DSH’s ctx.skills. The installation command on the directory page is dsh plugin add github:labring/sealos-skills. The source code and documentation are available at:
- Directory page: https://deepseek-harness-plugin.com/zh-CN/plugins/sealos-skills/
- GitHub: https://github.com/labring/sealos-skills