OceanBase Skills

A collection of AI Agent Skills for OceanBase products — designed for use with Claude Code, Cursor, Windsurf, and any other agent that supports the Agent Skills Specification.

Each skill is a self-contained directory with a SKILL.md file (plus optional references/) that gives AI agents the domain knowledge needed to help you operate OceanBase products correctly and safely.

---

Available Skills

oceanbase-deploy — OceanBase 部署与运维 (via obd)

Skill	Description
oceanbase-deploy	Overview & routing — start here if unsure which skill to use
cluster-management	Cluster lifecycle: deploy, start, stop, upgrade, scale out, OCP CE takeover, monitoring
tenant-management	Tenant CRUD, backup, restore, workload optimization
seekdb (obd)	obd-managed SeekDB: primary-standby HA (switchover / failover / decouple)
testing-and-benchmark	Sysbench, TPC-H, TPC-C, mysqltest benchmarks

seekdb — SeekDB 独立安装与编译构建

Skill	Description
seekdb	Overview & routing for standalone SeekDB
seekdb/install	Install/deploy SeekDB via Homebrew, Docker, yum, apt, pip, Windows MSI
seekdb/build	Build SeekDB from source for macOS, Linux, Android, Windows, Python wheel

More skills are on the way. Planned areas include OceanBase kernel tuning, SQL diagnostics, migration, and more.

---

Quick Start

Install via skills.sh (recommended)

# Install the oceanbase-deploy skill (includes all sub-skills)
npx skills add oceanbase/oceanbase-skills --skill oceanbase-deploy

# Install the seekdb standalone skill (install / build from source)
npx skills add oceanbase/oceanbase-skills --skill seekdb

# Or install all skills from this repo
npx skills add oceanbase/oceanbase-skills

Manual install — Claude Code

mkdir -p .claude/skills/oceanbase-deploy
curl -sL "https://raw.githubusercontent.com/oceanbase/oceanbase-skills/main/skills/oceanbase-deploy/SKILL.md" \
  -o .claude/skills/oceanbase-deploy/SKILL.md
for s in cluster-management tenant-management seekdb testing-and-benchmark; do
  mkdir -p .claude/skills/oceanbase-deploy/$s
  curl -sL "https://raw.githubusercontent.com/oceanbase/oceanbase-skills/main/skills/oceanbase-deploy/$s/SKILL.md" \
    -o .claude/skills/oceanbase-deploy/$s/SKILL.md
done

# SeekDB standalone skill (install + build)
mkdir -p .claude/skills/seekdb
curl -sL "https://raw.githubusercontent.com/oceanbase/oceanbase-skills/main/skills/seekdb/SKILL.md" \
  -o .claude/skills/seekdb/SKILL.md
for s in install build; do
  mkdir -p .claude/skills/seekdb/$s/references
  curl -sL "https://raw.githubusercontent.com/oceanbase/oceanbase-skills/main/skills/seekdb/$s/SKILL.md" \
    -o .claude/skills/seekdb/$s/SKILL.md
done

Load a single skill from GitHub URL

https://raw.githubusercontent.com/oceanbase/oceanbase-skills/main/skills/oceanbase-deploy/<skill-name>/SKILL.md

---

Agent Integration

Claude Code / Cursor / Windsurf (via skills.sh)

npx skills add oceanbase/oceanbase-skills --skill oceanbase-deploy

npx skills add 会自动检测你的 IDE（Claude Code、Cursor、Windsurf 等）并安装到对应目录。

Claude Code (manual)

Use the curl commands in Quick Start above, or manually place SKILL.md files into .claude/skills/oceanbase-deploy/.

Other Agents

• System prompt / rules file: paste the content of SKILL.md.
• URL-based loading: use the raw GitHub link above.
• In-conversation context: paste the SKILL.md content at the start of your session.

---

Example Prompts

After loading the skills, ask your agent for concrete tasks. Below are examples grouped by skill.

Cluster Management

部署一个本机 OceanBase 开源版本，能快速跑起来就行

用 config.yaml 部署一个名为 test-cluster 的 OceanBase 社区版集群

帮我部署 OCP

帮我直接启动 test-cluster，并检查启动后状态

如何给 ob-test 添加 Prometheus 和 Grafana 监控

Tenant Management

在 test-cluster 上创建一个名为 mysql 的租户

给 test-cluster 上的 mysql 租户配置备份路径并执行一次备份

SeekDB (obd-managed HA)

部署并启动一个 SeekDB 实例

创建一个 SeekDB 主备集群，并告诉我主库和备库分别怎么部署

查看 seekdb-test 的拓扑，如果主库挂了该用 switchover 还是 failover

SeekDB (standalone install / build)

在我的 Mac 上安装 SeekDB

用 Docker 部署一个 SeekDB 实例

帮我从源码编译 SeekDB 的 Linux rpm 包

Testing & Benchmark

对 test-cluster 的 mysql 租户跑一个 sysbench 测试

给我跑 TPC-H 的完整命令和参数

Tips

• Want the agent to execute directly? Say "帮我执行".
• Want a plan first? Say "先不要执行，只给我命令和步骤".
• For destructive operations, say "我确认允许高风险操作" or "先不要执行破坏性命令".

---

Repository Structure

oceanbase-skills/
├── .claude-plugin/
│   └── marketplace.json               # skills.sh discovery manifest
├── README.md
├── AGENTS.md
├── package.json
├── LICENSE
└── skills/
    ├── oceanbase-deploy/              # OceanBase obd deployment & ops
    │   ├── SKILL.md                   # Overview & routing
    │   ├── README.md
    │   ├── package.json
    │   ├── cluster-management/        # Cluster lifecycle
    │   │   ├── SKILL.md
    │   │   └── references/
    │   ├── tenant-management/         # Tenant ops
    │   │   ├── SKILL.md
    │   │   └── references/
    │   ├── seekdb/                    # obd-managed SeekDB HA
    │   │   ├── SKILL.md
    │   │   └── references/
    │   └── testing-and-benchmark/     # Benchmarks
    │       ├── SKILL.md
    │       └── references/
    └── seekdb/                        # Standalone SeekDB
        ├── SKILL.md                   # Overview & routing
        ├── package.json
        ├── install/                   # Install via Homebrew/Docker/yum/apt/pip/MSI
        │   ├── SKILL.md
        │   └── references/
        └── build/                     # Build from source (macOS/Linux/Android/Windows)
            ├── SKILL.md
            └── references/

Each skill follows the Agent Skills Specification:

File	Purpose
SKILL.md	Main skill content with YAML frontmatter — consumed by AI agents
references/*.md	Supplemental documentation, loaded on demand to save context

---

Contributing

Contributions are welcome. To add a new skill:

1. Create skills/<your-skill-name>/.
2. Write SKILL.md with YAML frontmatter (name, description, compatibility, metadata) and clear, agent-friendly instructions.
3. Add references/ for detailed supplemental content (loaded on demand, not always in context).
4. Keep SKILL.md under 500 lines; move details to references/.
5. Test your skill by loading it into Claude Code and running representative prompts.

Please keep skill content factual, concise, and safe — especially for destructive or irreversible operations.

---

License

MIT