Case Study: Automatic Requirement Filing and Notifications with Claude Code + lovrabet + Lark CLI

Scenario

On a machine with the Lovrabet CLI installed, a single sentence to Claude Code — for example, "File a requirement for Zixuan: get the page-editing guide into the help center as soon as possible" — and Claude Code automatically:

1. Identifies the owner
2. Creates the requirement in the Lovrabet system
3. Sends an email notification to the owner

No manual steps along the way.

---

What you need before starting

1. A Lovrabet account

If you don't have one yet, register at the Lovrabet User Center first.

2. An app with requirement management enabled

You need a Lovrabet app that has completed database reverse engineering. This case uses our own Qizhi Yuntu Project Management app as the example (appcode: app-173e8652).

If you use your own app, just replace app-173e8652 and the related dataset codes in this guide with yours.

3. A Mac or Linux machine

Everything here was verified on a Mac; the commands are identical on Linux.

---

Step 1: Install the CLI tools

Open a terminal and run the following commands in order.

1.1 Install the lovrabet CLI and Lovrabet skills

npm install -g lovrabet
npx skill add lovrabet/lovrabet-cli -g -y

1.2 Install lark-cli (the Lark CLI)

npx @larksuite/cli@latest install

1.3 Install the Claude Code CLI or another Agent client

Install whichever Agent IDE you prefer, following its official documentation — we won't repeat that here. This guide uses Claude Code to walk through the whole process.

Once installed, verify everything is in place:

lovrabet --version
lark-cli --version
claude-code --version

---

Step 2: Sign in and authorize

2.1 Sign in to lovrabet

In the terminal, run:

lovrabet auth login

Follow the prompts to enter your Lovrabet account and finish signing in.

2.2 Verify that the app list loads (using Qizhi Yuntu Project Management as the example)

# 查看已注册的应用列表
lovrabet app list --format json

What is an appcode? Every Lovrabet app has a unique identifier called the appcode — essentially the app's "name". You use it to tell the CLI which app to operate on.

2.3 Sign in to Lark (lark-cli)

lark-cli auth login

This opens a browser; follow the prompts to authorize with your Lark account, then return to the terminal.

---

Step 3: Start using it

3.1 Launch Claude Code

Open a terminal in any directory and type:

claude

3.2 State the request

Say it in one sentence and Claude Code completes the entire flow. For example:

/lovrabet 给子轩提个需求：尽快把页面修改指引放到帮助中心并配置入口。
补充信息：
1）从成员表查出负责人信息
2）根据标题关键词匹配项目
3）把需求写入对应的项目管理系统
4）发邮件通知负责人,邮件正文要加入需求描述、项目管理系统对应需求详情页的网址并且可点击。
5) 执行完后返回需求 ID 和邮件状态

3.3 Wait for execution

Claude Code automatically:

• Looks up Zixuan's project member information
• Creates the requirement (priority: high, status: new)
• Emails Zixuan a notification

When it finishes, you'll see the requirement ID and the email delivery result.

---

Step 4: Capture the flow as a Skill (optional)

If your team files requirements often, letting the AI improvise from scratch every time is both tedious and inconsistent in quality. You can capture the flow you just ran as a reusable Skill.

4.1 Why capture it as a reusable skill

• You notice a business flow repeating frequently
• You want to codify internal know-how so the AI follows it instead of "re-understanding" it each time
• You want teammates to trigger the same flow with one sentence — no need to know the business process, just reuse the accumulated experience

4.2 How to do it (using requirement filing as the example)

After running the full flow once in Claude Code, just say:

 用skill-creator 这个skill 把刚才的操作封装成一个 业务Skill，方便以后复用

The AI analyzes what it just did, which tools and parameters it used, then calls skill-creator to generate the Skill files. When it's done, it tells you the Skill's name and where it lives.

Suppose the generated Skill is named: project-issus

4.3 How to use it afterwards?

From then on, you only need to say:

帮我用「project-issus」给子轩提个需求：App背景色可以根据用户手机壳颜色自动变色

The AI loads the Skill automatically and follows its steps — no need to describe the flow again.

---

FAQ

Q: The claude command doesn't respond?

Make sure the Claude Code CLI is installed:

claude-code --version

If you get command not found, reinstall:

npm install -g @anthropic-ai/claude-code

Q: lovrabet auth login fails?

Confirm you're using a registered Lovrabet account and that the bound user AccessKey is correct.

Q: lark-cli auth login says insufficient permissions?

Your Lark account probably lacks the required app permissions. Ask your Lark admin to grant access to the Lark app.

Q: How do I view the requirements that were created?

In the project management app you created in Lovrabet. The CLI exists so that Agents operate only Lovrabet-built apps, simulating human operations — so all results live inside the app itself.

---

Technical architecture (optional reading)

The solution uses a three-layer architecture:

Layer	Tool	Responsibility
Orchestration	Claude Code + global Skills	Understand intent, orchestrate the flow, map members
Data	lovrabet CLI	App switching, data CRUD
Delivery	lark-cli	Send email notifications

Each layer does its own job, and the Skill glues them into one complete business scenario.

---

Last updated: April 2026