🦩 Foreword

In daily life, we've all encountered this situation more or less:

You chat with an AI for half an hour, and it produces a well-reasoned plan. But then you still have to open Feishu (Lark) yourself, paste the content in word by word, manually create tables, manually send messages, and manually schedule meetings.

The AI said a lot, but it didn't actually do anything for you.

People often think the AI isn't smart enough, but the reality is, you haven't given it a pair of capable hands.

What we're going to do today is give the agent inside Feishu a pair of hands—using Codex + Feishu CLI, we'll enable AI to truly work for you inside Feishu: writing documents, building multidimensional tables, sending messages, scheduling meetings, processing emails... You just talk, and it does the work.

1. 🦌 First, Understand: What is Feishu CLI

Before we get our hands dirty, take two minutes to understand one thing: What problem does Feishu CLI actually solve?

Previously, there was no communication channel between AI and Feishu. You'd ask AI to help you write meeting minutes; it would finish writing, but you'd still have to copy and paste them into a Feishu document yourself. The AI gave you 'advice,' but couldn't help you 'execute.'

Feishu CLI (lark-cli) is an officially open-sourced command-line tool by Feishu. What it does is pull Feishu 'into' the AI's workflow. Once installed, an AI Agent (like Codex) can directly:

• Create and edit cloud documents in Feishu
• Read and write data in multidimensional tables
• Query and create calendar events
• Send messages, reply in group chats
• Process emails, manage tasks
• Operate slides, whiteboards, knowledge bases...

It covers a total of 11 business domains and 100+ skills, under the MIT open-source license, with no application or approval needed—available to everyone.

To put it in plain terms: Feishu CLI isn't about teaching you the command line; it's about letting the AI use the command line to operate Feishu for you. You just need to install it, and then continue talking to the AI in natural language.

2. 🦒 Step-by-Step: Installing Feishu CLI in Codex

1. Prerequisites

Before starting, make sure your computer has Node.js version 16.0 or higher installed.

Not sure? Open your terminal (Terminal on Mac, PowerShell on Windows) and enter:

node -v

If you see a version number (e.g., v20.0.0), it's already installed. If it says 'command not found,' go to nodejs.org to download and install it.

2. Install Codex

Open this link: https://chatgpt.com/zh-Hans-CN/codex/, and download the Codex desktop version.

3. Install Feishu CLI

Open Codex and send it the following message directly:

Please help me install the Feishu CLI tool and all related Feishu Skills. The specific steps are as follows:
Step 1: npm install -g @larksuite/cli
Step 2: npx skills add https://github.com/larksuite/cli -y -g
Step 3: lark-cli config init --new
Tell me the result after installation is complete.

After installation, this will be displayed👇🏻:

*A key point to note here is: *Don't just say 'help me install Feishu CLI,' as this might only install the tool itself without the Skills. Use the complete prompt above to ensure all 19 Feishu skills are installed.

4. Configure the App and Authorize

(1) Configure the App

After installation, Codex will guide you through the app configuration. This step will automatically create a Feishu app (or you can choose an existing one), without needing you to manually operate in the developer platform.

Next, Codex will redirect you to a webpage to create a Feishu CLI app. Just follow the steps one by one.

Click the 'Create' button to finish creation.

Finally, your Feishu client will send you a message like this, indicating that your CLI app has been created successfully.

(2) Complete Authorization

After configuration, run the authorization command to allow the AI to operate Feishu on your behalf:

lark-cli auth login
lark-cli auth login --recommend

After execution, a link will be generated. Copy it and open it in a browser, then scan the QR code in Feishu to confirm.

The authorization list is shown below👇🏻:

Explanation of Two Modes:

• Without Authorization: The AI can create documents, send messages, etc., but cannot read your personal data (calendar, private messages, etc.)
• With Authorization: The AI can access your personal calendar, messages, and documents on your behalf, providing more complete functionality.

5. Restart Codex and Verify Installation

This step is crucial—after configuration, you must restart Codex for the Skills to load completely.

After restarting, send a test command:

Help me create a cloud document in Feishu, with the title "Feishu CLI Test Document" and the content: Feishu CLI installed successfully, AI is in position.

If this document appears in your Feishu, congratulations—your agent has grown hands and feet.

3. 🦙 8 Practical Scenarios: Let AI Finish the Job for You

Once installed, let's see what this 'winged agent' can actually do. Below are 8 of the most practical scenarios, each with a prompt you can copy and use directly.

1. Cloud Documents: AI Drafts, You Review

It's 3 PM on a Friday, and your boss suddenly says, 'I need a Q3 review document for Monday.' You open the AI and start chatting. Twenty minutes later, you have a framework and content, and then... you start manually copying the content into Feishu piece by piece. The heading formats are wrong, the pasted tables are a mess, and you spend another half hour cleaning it up.

This feels like having a chef design a recipe for you, and then you have to cook it yourself.

Now, with Codex + Feishu CLI, how do you do it?

Directly ask the AI to create the document in Feishu. You can use comments in the document to suggest revisions, and the AI will read the comments and modify the text, all without leaving Feishu.

(1) Project Review

Prompt Example:

https://jwf86xh1hew.feishu.cn/wiki/UGRjwXk4eiZ649kPMhEckqK0nxe?from=from_copylink

Above is the meeting minutes our team discussed this afternoon. Based on what we discussed, help me create a cloud document in Feishu with the theme "Q3 Project Review."
The document structure should include: Project Background, Core Achievements, Problems Encountered, Next Phase Plan.
First, write a framework with 2-3 key points for each section.

Cloud document creation display👇🏻:

(2) Let AI Be Your Reviewer

Often, you let AI write a first draft, but after it's done, you don't know where the problems are?

Throw the document link to the AI and ask it to review it as a 'picky senior colleague'. Don't let it modify directly, just leave comments on problematic areas—logical loopholes, vague expressions, missing data—all marked for you.

{{Document Link}} 
Read this document and review it as a senior colleague.
Do not modify it directly. Just use word comments to mark the problematic areas,
and write specific modification suggestions in the comments.

The original first draft of the article looked like this:

Let's look at the Codex task execution process:

After execution, reopen the link. You'll find several fresh comments have appeared👵

We can see that the AI's comments appear directly in the Feishu document. You can accept or reject them one by one, fully controlling the final content. The whole process requires no window switching, just like having an always-online colleague sitting next to you helping to edit.

2. Multidimensional Tables: Data Entry and Querying, Done in One Sentence

Every Monday morning, you open Feishu, look at that dozens-of-rows project progress table, and start updating statuses, filling in deadlines, and changing assignees row by row. This task itself has no difficulty, but it takes you twenty to thirty minutes, and your brain is idling the whole time.

What's worse is when someone suddenly asks, 'What high-priority tasks are still unfinished?' You have to manually flip through and filter, searching for a while.

So now, with Feishu CLI, what can we do?

Describe the operation you want in natural language, and the AI will directly help you build tables, enter data, and make queries.

(1) Create a Table

Prompt Example (Create Table):

Help me create a multidimensional table in Feishu to track this week's project tasks.
Fields include: Task Name, Assignee, Deadline, Priority (High/Medium/Low), Status (Not Started/In Progress/Completed).
First, help me enter 10 sample data records.

You just need to state your requirements, and Codex will help you create the corresponding multidimensional table:

Finally, let's see the created effect:

(2) Query Data

Prompt Example (Query Data):

Check the [Task Tracking Table] for all tasks where the status is "In Progress" and the priority is "High."
List the task names and assignees, sorted by deadline.

Codex will directly help you view high-priority and in-progress tasks:

At the same time, it thoughtfully created a view for us. This makes it convenient for future direct viewing by clicking the view:

(3) Make the Table 'Run by Itself'

After building the table, you can also have the AI configure an automation rule for you. For example: if a task's deadline arrives but its status is still 'In Progress,' automatically send a Feishu message to remind the assignee; or when a new task is created, automatically notify the project group. These used to require manual step-by-step configuration in Feishu's automation, but now you can just say it in one sentence👇🏻:

Help me configure an automation in the [Task Tracking Table]:
When a record's "Deadline" arrives on the current day, and the "Status" is still "In Progress,"
automatically send a Feishu message to the record's "Assignee" with the content:
"Reminder: You have a task [Task Name] due today. Please update the progress promptly."

Automation configuration effect:

AI doesn't just help you 'build' tables; it can also help you 'use' them and make them 'run by themselves'—querying data, updating statuses, batch modifications, configuring automation reminders, all handled in natural language. That twenty to thirty minutes of Monday routine operations can be saved from now on, and you don't need to manually monitor approaching deadlines.

(4) Cross-Table Aggregation

Cross-table aggregation, a weekly report in one sentence:

The project team has three tables: Task Tracking Table, Bug Record Table, and User Feedback Table. Every Friday, you need to summarize the key data from these three tables into a weekly report for your boss.

The old way was to open each one, manually copy, and piece it together into a document, taking about forty minutes. Now, just tell the AI what you want. It reads across tables, summarizes, and writes it into a document in one step.

Just tell Codex:

Help me summarize this week's project progress. Data comes from three multidimensional tables:
1. [Task Tracking Table]: Count the number of new tasks, completed tasks, and overdue tasks this week.
2. [Bug Record Table]: Count the number of new bugs, fixed bugs, and unresolved P0-level bugs this week.
3. [User Feedback Table]: Count the number of new feedback items this week, and list summaries of feedback with a rating below 3.

Write the summary into a Feishu document titled "Week X Project Weekly Report."

Format: Each of the three modules has a subheading. Data should be bolded, and anomalies should be marked in red.

Finally, Codex will generate a project weekly report based on the data in your past multidimensional tables:

The generated weekly report effect is as follows:

The AI simultaneously reads data from multiple tables, automatically performs addition/subtraction statistics, and then writes the results into a document in a formatted way. Cross-table aggregation, which previously required manual 'brick-moving,' is now a matter of one prompt. The weekly report goes from 'the most annoying thing' to 'just a glance before sending it out.'

(5) Lightweight Dashboard

Turn a multidimensional table into a lightweight data dashboard.

Multidimensional tables have a rarely used feature: Dashboard View. It can directly render data from the table into charts—bar charts, line charts, pie charts, number cards—without needing to export to Excel or manually create charts. When the data updates, the charts automatically follow.

Previously, configuring a dashboard required dragging components, selecting fields, and adjusting styles one by one in the Feishu interface, taking over half an hour for a decent dashboard. Now, just say:

Help me create a dashboard view in the [Task Tracking Table] named "Project Progress Dashboard."
Include the following four chart components:
1. Number Card: Display the total number of tasks currently "In Progress."
2. Pie Chart: Show the distribution of all tasks by "Priority" (High/Medium/Low).
3. Bar Chart: Group by "Assignee" and show the number of tasks per person.
4. Line Chart: Show the trend of task completion count by week based on "Deadline" (last 4 weeks).
Overall layout in two columns, with the number card in the top left corner.

Codex dashboard generation process:

A super informative dashboard is ready~

Once the dashboard is built, every time you open the multidimensional table, just switch to the 'Project Progress Dashboard' view to see all key metrics on one screen. When data changes, there's no need to reconfigure; the charts refresh automatically. This approach is especially suitable for PMs and project leads who need to regularly report progress upwards—take a screenshot of the dashboard, and the weekly report is done.

3. Slides: From Idea to Presentation

You have to give a department presentation tomorrow, and you only remember to make the PPT tonight. You open Feishu Slides, stare at the blank page for five minutes, not knowing where to start. After you figure out the outline, you spend two hours laying out pages, adjusting fonts, and aligning elements, and you haven't even filled in half the content yet.

The truly valuable part of making a PPT is 'figuring out what to say,' but most of the time is consumed by layout and formatting.

So now, with Codex × Feishu CLI, how do we handle it? Tell the AI your topic, audience, and approximate page count, and it will directly generate an editable slide framework in Feishu. You just need to fill in the content and tweak the details on top of it.

(1) New Employee Onboarding PPT

Prompt Example:

Imagine a scenario: You are the HR at an AI hardware startup that just completed Series A funding. Next Monday, you need to conduct onboarding training for 8 new hires. The team is expanding rapidly, and there are no ready-made training materials. It's 4 PM on Friday.

Previously, you could only throw together a makeshift PPT. Now, you directly tell the AI:

I am the HR of an AI hardware startup that just completed Series A. The team expanded from 30 to 50 people.
Next Monday, I need to conduct onboarding training for 8 new colleagues, and there are no existing materials.
Help me create an onboarding training slideshow in Feishu, about 10 pages.

Company Info:
- Company Name: Jiuyuan Quadrant
- Direction: AI wearable devices (main product is a smart headphone with real-time emotion perception)
- Stage: Series A, Pre-revenue, currently polishing the product and building the sales team
- Location: Wangjing, Beijing, occasional remote work possible

Content Structure:
Page 1: Welcome page (Company name + one-sentence product introduction)
Page 2: What we do (Product direction + problem solved + target users)
Page 3: Current company stage (Series A completed, core goals for the next 6 months)
Page 4: Which team you'll join (Product, Hardware R&D, Software R&D, Marketing, HR)
Page 5: New hire first-week checklist (Equipment pickup, Feishu permissions, product experience, 1-on-1 alignment with direct leader)
Page 6: Company culture (Three words: Fast, Direct, Lean—act quickly, give direct feedback, cut the fluff)
Page 7: Common tools (Feishu for collaboration, Notion for product docs, Linear for task management, Figma for design drafts)
Page 8: Probation period rhythm (Month 1: integrate, Month 2: get hands-on, Month 3: deliver output)
Page 9: FAQ (When are stock options granted? Can I bring my own device? How to reimburse?)
Page 10: Closing (Welcome aboard, let's build this product together)

Style: A bit cool, unpretentious, direct. Light color scheme. No more than 4 key points per page.

Codex will use the lark-slides function to create the Slides:

Finally, it will directly create the training slides for this hardware company in your Feishu:

Let's take a look at the slideshow effect:

What the AI generates are slides that can be directly edited in Feishu, not screenshots or PDFs. The framework, text, and page structure are all there. You only need to do the final 20% of adjustments, instead of building that 80% scaffolding from scratch.

(2) Fashion Ordering Fair

Another scenario: Fashion Ordering Fair.

Twice a year, you have to present a new collection launch to ordering agents and buyers from all over the country. This presentation is one that needs to make them open their wallets and place orders—the brand story must be clear, the outfit combinations understandable, and the profit margins calculable.

Previously, making this Slide deck required designers to produce images, the product team to provide data, and you to piece together the content page by page, aligning back and forth for three or four days, and it still might not look good in the end.

Now, you tell the AI the brand tone and content structure, and it directly generates an editable ordering fair Slide deck in Feishu. You just need to fill in the real product images and data.

Generate an 8-page high-interaction Slide deck for the emerging fashion brand's 2026 Spring/Summer New Arrivals Ordering Fair.
Brand Tone: Morandi color palette, minimalist tailoring, high-end sensual cool style.
Audience: Ordering agents/buyers from various regions, concerned with brand story, outfit combinations, and profit margins.

Fashion Ordering Fair PPT effect as follows:

4. Whiteboard: Charts Directly Become Editable Vector Graphics

The team is discussing a business process, and you use AI to help you draw a flowchart. The Mermaid code is beautifully written. Then you want to put it into Feishu for everyone to see—a screenshot? The resolution isn't good enough. Copy the code? Colleagues won't understand. Redraw it manually in the Feishu whiteboard? That would take forty minutes.

In the end, you paste a screenshot, and a colleague says, 'I can't see clearly where this arrow points,' leaving you to explain verbally.

But now, we can: Directly tell the AI what process you want to draw, and it will generate editable vector graphics in the Feishu whiteboard—not a screenshot, but a truly live diagram where you can click in to change nodes, colors, and add content.

(1) Flowchart - New Employee Onboarding

Prompt Example:

Help me draw a new employee onboarding flowchart in the Feishu whiteboard:
HR sends offer → Candidate signs contract → On the start date, receive equipment + create accounts 
→ First week shadowing → Probation review (Pass/Fail) → Confirmation or departure.
Use different colors to distinguish Pass and Fail. The overall style should be clean and suitable for pasting into a Feishu document for new hires to see.

Codex will generate a very fitting onboarding flowchart for you:

Mermaid / PlantUML charts are automatically converted to Feishu whiteboards. What's generated are vector graphics, not pasted images. If you need to change the process, add nodes, or adjust colors later, you can operate directly in the whiteboard, and colleagues can also continue editing and collaborating on it. This capability is quite stunning within Feishu CLI.

The whiteboard's playability goes far beyond just drawing a flowchart. The following four tasks increase in difficulty progressively. Interested friends can try them out on their own after class.

(2) Organizational Chart - Business Management

Draw the organizational chart ⭐

The company just completed a round of hiring expansion, and HR needs to update the org chart.

The old way was to open PPT, manually drag nodes, spend half an hour aligning, and then find out someone was missed and start over.

Now, directly tell the AI the structure, and it will generate an editable organizational chart in the Feishu whiteboard—if personnel change, just modify the nodes directly, no need to re-screenshot, and colleagues can continue collaborating on it.

Help me draw a departmental organizational chart in the Feishu whiteboard.
Company Name: Jiuyuan Quadrant
Structure as follows:
Under the CEO, there are four lines:
- Product Line: Product Director → Product Manager×2, User Researcher×1
- R&D Line: CTO → Hardware R&D×3, Software R&D×4
- Marketing Line: Marketing Director → Branding×1, Growth×2
- Functional Line: HR×1, Finance×1
Use placeholders for names (e.g., "Product Manager A"). Use rounded rectangles for boxes. Differentiate levels with different colors.
The top-level CEO node should have a dark background + white text. Use different main colors to distinguish each line. Overall vertical layout.

Finally, Codex generated a whiteboard in Feishu:

Add another sentence:

Insert this diagram into the Feishu document "Jiuyuan Quadrant Team Introduction"
https://jwf86xh1hew.feishu.cn/wiki/Td5lw9Xhei7nokked38cpmrdnqe

The AI will directly embed the whiteboard into the document, without you needing to manually copy the link. For example, below is the original empty document:

Then Codex directly exported the flowchart from the whiteboard as an image and inserted it into the new document:

(3) Mind Map - Deconstructing the Problem of Low User Retention

For the third whiteboard scenario, we use a mind map to deconstruct a complex problem.

In daily work, sometimes before a meeting, we might need to clarify our own thoughts. Or, after completing a competitive analysis, we want to organize it into a diagram.

At times like these, we can have the AI directly generate a mind map in the whiteboard, which is much more intuitive than listing points in a document.

The key to this task isn't 'drawing the diagram,' but forcing yourself to clearly deconstruct the problem before giving it to the AI—the clearer the structure you provide, the more useful the diagram the AI draws.

For example, we can give a prompt like this:

Help me create a mind map in the Feishu whiteboard with the theme "Why is our user retention rate low."
Starting from the central node, expand into 4 directions: Product Experience, User Onboarding, Competitive Impact, Pricing Factors.
List 2-3 specific reasons under each direction. Use small rounded rectangles for reason nodes, with a lighter color shade than the parent node.
Use thick lines (3px) for main branches and thin lines (1px) for sub-branches. The overall layout should expand left and right, with the central node centered and bolded.

Finally, Codex generates a very comprehensive mind map:

After the mind map is drawn, you can continue to have the AI organize the content from the diagram into an outline for a Feishu document.

Two steps to complete the full process of 'think clearly → write clearly.'

(4) Whiteboard and Document Linkage

First, have the AI read the content inside a Feishu document, and then automatically generate a corresponding whiteboard diagram based on that content, creating a true linkage between the two files.

Prompt Example:

Read this document: {{Paste the Feishu document link here}}
Based on the "Q3 Project Timeline" section in the document,
generate a Gantt-chart-style timeline in the Feishu whiteboard.
The horizontal axis is months (July-September), and the vertical axis is each sub-project.
Use green for completed nodes, blue for in-progress ones, and gray for those not yet started.
Label each node with the project name and start/end times.

Gantt Chart Effect Display:

In this scenario, the main point is to let the AI first help you learn, and then draw the flowchart.

(5) High-Density Information Graphics

Driven by JS scripts, generate high-density information graphics ⭐⭐⭐⭐

This scenario can be understood as the true ceiling of the whiteboard's capabilities.

For the first four tasks, the AI used DSL or Mermaid syntax—essentially describing 'what nodes exist and how to connect them.'

In this case, it uses JavaScript script-driven layout. The AI will directly write a piece of JS code to precisely calculate the coordinates, size, and color of each element, and then render it onto the whiteboard.

This means you can generate diagrams that are 'impossible to draw manually.'

Scenario A: Product Version Roadmap

Take the product's iteration history from v0.1 to the present and turn it into a horizontal timeline with alternating top-and-bottom cards. Clearly write the release time and key capabilities for each version—suitable for putting into a Q3 review PPT or an investor deck.

Draw a product version roadmap for the Jiuyuan Quadrant smart headphone in the Feishu whiteboard.
The timeline is arranged horizontally, from left to right:
- v0.1 (October 2025): Hardware prototype, only supports heart rate collection
- v0.2 (December 2025): Added galvanic skin response sensor, emotion recognition v0.1 online
- v0.3 (March 2026): App prototype released, Bluetooth connection stable
- v0.5 (July 2026): EVT validation passed, algorithm accuracy 78%
- v0.8 (September 2026): iOS beta version launched on TestFlight
- v1.0 (Q4 2026, planned): App Store launch, 200 seed users activated

Version nodes alternate top and bottom (odd numbers on top, even numbers on bottom).
Use solid cards for released versions, and dashed borders + light fill for planned ones.
Each card should state the version number, time, and 1-2 key milestones.
Overall style: Light background, main color green, white font.

The final generated product version roadmap effect is as follows:

---

Scenario B: Competitive Comparison Radar Chart + Description Card

Turn the results of a competitive analysis into a comparison chart that can be directly pasted into a PPT—not a table, but an infographic with visual weight.

Generate a competitive comparison infographic in the Feishu whiteboard with the theme "AI Emotion-Sensing Wearable Device Market."
Comparison dimensions: Emotion Recognition Accuracy, Hardware Comfort, App Experience, Price Range, Open API.
Products: Jiuyuan Quadrant (Us), Competitor A (Muse S), Competitor B (Emotiv EPOC), Competitor C (Garmin HRV).
Use a radar chart to display scores for each dimension (out of 10), one line per product, distinguished by color.
Place a description card to the right of the radar chart listing our core differentiators:
1. The only consumer-grade headphone form factor focused on emotion sensing
2. Algorithm open SDK, supports third-party integration
3. Price band positioned at 800-1200 RMB, lower than medical-grade competitors
Overall style: White background, clean color scheme, suitable for direct screenshot insertion into an investor deck.

The final generated radar chart effect is as follows:

---

PS: This prompt carries a lot of information. The AI's runtime will be a bit longer than the first three tasks (usually 3-5 min). This is normal—it's running JS scripts to calculate coordinates in the background. Be patient, the result is usually well worth it.

5. Spreadsheets: That Thirty Minutes of Mechanical Operation Every Week, No More

At the end of every month, you have to open that sales data sheet, paste in a new batch of data, manually calculate completion rates, mark overachievers in red, fix incorrect formats, remove spaces from names... all repetitive actions. Your brain doesn't need to engage the whole time, but it still takes you half an hour.

After finishing, you feel a strange kind of fatigue—not because it was difficult, but because it was boring.

How to do it now: Describe the data operations you want in one paragraph, and the AI processes it directly. You can go get a glass of water and come back when it's done.

Prompt Example:

Open the [Sales Data Sheet], uniformly format all names in column A to "Last Name First Name" (add a space between last and first name),
then add a "Completion Rate" field in column E, calculated by dividing column C (Actual Sales) by column D (Target Sales).
Keep the result to two decimal places, and mark those exceeding 100% in red.

Let's see the before-and-after effect comparison:

The AI can directly translate your natural language descriptions into spreadsheet operations. Those monthly routine data tidying tasks can now be delegated with one paragraph, letting it run the rest.

6. Instant Messaging: Mass Notifications, but Each Feels Individually Written

For the sixth scenario, let's talk about instant messaging.

You've been following up on a project that has reached a key milestone. You need to notify over a dozen relevant colleagues, each with slightly different situations—some need to confirm a time, some need to submit materials, some just need a heads-up.

Sending them one by one, slightly tweaking the content each time, makes your hands sore by the end. Or you could send one mass message, but mass messages always feel less personal and are easily ignored.

How to do it now?

Tell the AI who to notify and what to convey to each person. It automatically searches the address book and sends messages one by one, personalizing each message based on the recipient's situation.

Prompt Example:

Help me send a Feishu message to each of the following colleagues, telling them there's a new project brainstorming meeting this Friday at 3 PM,
and asking them to confirm if they can attend: Yake, Haibao, Yanyu, Ajing.
The message should include the meeting theme "Q3 Content Topic - Brainstorming Session," in a friendly yet formal tone.

The AI will first ask you to add an authorization, allowing it to send a message on your behalf:

Finally, the AI successfully used my identity to send messages to my four colleagues separately✌🏻:

In this scenario, we can see that the AI automatically finds the corresponding people in the address book and sends messages one by one. For a dozen messages, you only need to say it once, and it finishes sending them. For scenarios requiring regular team member notifications, this feature can completely strike a 'not hard but annoying' task from your list.

7. Calendar: Scheduling Shouldn't Rely on Back-and-Forth Messaging

For the seventh scenario, let's talk about the calendar.

Suppose you need to schedule a meeting with four people. You ask A first; A says Wednesday works. You ask B; B says Wednesday afternoon is busy, how about Thursday? You ask C; C says they're on a business trip Thursday morning. You ask D... After twenty minutes of back-and-forth, it's still not scheduled, and you postpone it to next week.

If the team spans time zones, this becomes even more painful—you have to manually calculate time differences and guess what counts as 'reasonable working hours' for the other person.

How to do it now?

Tell the AI who to invite and what the meeting is about. It directly pulls everyone's calendars, finds common free slots, considers time zones, recommends a few options, you pick one, and the meeting is created.

Prompt Example:

Help me check the calendars for Yake, Haibao, and Carol this week.
Find a time slot when everyone is free for a 1-hour project discussion meeting.
Preferably between 10 AM and 5 PM on a workday morning.
After finding a suitable time, directly help me create the meeting with the title "Q3-Q4 Planning Discussion."

I added two Carols, and the AI first asked me which Carol I meant:

After confirming which Carol, the AI helped me complete the calendar event creation:

Finally, in my calendar, I saw the created event✌🏻

In this scenario, we can see that the AI pulls each person's calendar data, automatically calculates common free time, recommends options, and then directly creates the meeting. That twenty minutes of back-and-forth is compressed into the time it takes you to say one sentence. It also handles cross-time-zone issues, so you don't need to open a world clock to convert.

8. Tasks: After the Meeting, Don't Let Action Items Die in the Group Chat

You've definitely had this experience: After an hour-long meeting, everyone says, 'Okay, so that's settled,' and then disperses. Three days later, someone asks, 'Who was following up on that thing we talked about last time?' Everyone looks at each other blankly, scrolls through the group chat for a while, and finds the sentence that was said back then—but no one created a task, no one set a deadline, and the matter just quietly disappeared.

It's not that people are irresponsible; it's that at the end of a meeting, everyone is tired, and no one wants to open the task system to enter items one by one.

But now, by combining Codex × Feishu CLI, we can:

After the meeting, throw the Minutes link to the AI. It automatically scans the full text, extracts all mentioned action items, creates them one by one in Feishu Tasks, marks the assignee and deadline, and finally sends the list to the group. You don't need to do anything in this entire process.

Prompt Example:

Read the meeting minutes from this afternoon's 'Summer Ordering Fair'.
Extract all the action items mentioned in it.
Create them one by one in Feishu Tasks, marking the assignee and deadline for each (if mentioned in the minutes).
After creation, send the task list to the 'Summer Fashion Ordering Fair Project Team' group.

The original 'Q2 Ordering Fair Review Meeting' minutes are as follows:

The AI will directly help me summarize the TODO items and send them to the conversation:

It's clear that between the end of a meeting and the landing of tasks, there used to be a gap that relied on 'personal conscientiousness.' Now that gap is filled. The AI extracts from the Minutes, directly creates tasks, and notifies the relevant people. Everyone can see their own to-dos as soon as they open Feishu, without needing to scroll through group messages to find 'what was said back then.'

Alright! Through these eight major scenarios, I believe everyone has felt the joy of Feishu CLI 🥳🥳🥳

4. Capability Panorama: Which Business Domains Does Feishu CLI Cover

Finally, let's use a table to make an overall summary of the capability panorama of Feishu CLI:

Business Domain	Core Capabilities
💬 Messages & Groups	Search messages and group chats, send messages, reply to topics
📄 Cloud Docs	Create docs, read content, update text, comment collaboration
🗂️ Cloud Space	Upload/download files, manage permissions, process comments
📋 Spreadsheets	Create sheets, read/write cells, batch update
📊 Multidimensional Tables	Manage data tables, fields, records, views, dashboards, automation
📅 Calendar	Query events, create meetings, query availability, recommend times
🎥 Video Meetings	Search meetings, get minutes and transcripts, link to calendar docs
📧 Mail	Search, read, draft, send, reply, archive emails
✅ Tasks	Create tasks, update status, manage lists and subtasks
📚 Knowledge Base	Query spaces, manage nodes and document hierarchy
👥 Contacts	Query users, search colleagues, view departments

5. 🦘 FAQ & Pitfall Guide

Here are some Frequently Asked Questions & Pitfall Guides for everyone. Feel free to come back and check when you encounter problems 😋

No.	Problem	Solution
01	After installation, it says 'command not found'?	Confirm the CLI directory is added to the system PATH. Execute npm root -g to see the global directory, and add that path to PATH.
02	Configured, but AI seems to lack Feishu capabilities?	You must restart Codex! Skills are not hot-reloaded; they only load fully after a restart.
03	Authorization failed, saying 'authorization code expired'?	The OAuth authorization code is only valid for a few minutes. Re-execute lark-cli auth login.
04	Insufficient permissions error when calling?	Supplement authorization based on the error message: lark-cli auth login --scope "<missing_scope>". The CLI will tell you what permission is missing and provide an application link.
05	Can the enterprise admin control permissions?	Yes. The app created by the CLI still follows the enterprise's unified governance rules. Admins can manage app permissions centrally in the Feishu developer platform backend.
06	Does it support the international version of Lark?	Yes. Configure the international version of Lark's app via lark-cli config init.

6. 🦢 Conclusion

It's been less than two or three months since Feishu CLI was open-sourced, and it has already unlocked 100+ office skills. This speed itself says one thing: The process of AI taking over office tools is much faster than we imagined.

What we did today wasn't just installing a tool. We are redefining what it means to 'use Feishu'—from 'I open Feishu and operate it manually' to 'I say one sentence, and the AI finishes the tasks in Feishu for me.'

For professionals, this means a large amount of repetitive information transfer, format tidying, data entry, and message notifications can be struck off your work list. Your time and energy can be reserved for tasks that truly require judgment and creativity.

Install Codex + Feishu CLI, and starting today, let your Feishu truly grow hands and feet.

That's all for this article. See you next time 🍻🍻🍻