Feishu CLI Prompt Book (Best Practices Collection)
A curated collection of prompts and best practices for Feishu (Lark) CLI, helping you turn AI into a real productivity assistant. Every prompt is ready to copy — just replace the values inside 【】 with your own.
1. Meetings & Minutes (Miaoji)
After the meeting, action items are auto-assigned to everyone
Read this meeting transcript [Minutes URL] and extract every action item mentioned. Create a task for each owner, then send a notification message to everyone in the [Group Name] chat.No more back-and-forth scheduling
Help me schedule a one-hour sync next week with everyone in [Group Name]. Account for time zones and recommend 3 time slots that work for all. Once I confirm, create the meeting and pull people into a chat.Auto-generate meeting minutes as a Feishu Doc
Read this transcript [Minutes URL] and produce meeting minutes including key discussion points, decisions, and action items. Then create a Feishu Doc titled with today's date plus the meeting name.Last week's 8 meetings, decisions on one dashboard
Read all my meeting transcripts from last week. Extract decisions, open questions, and action items into a Feishu Base (three columns), then generate a simple dashboard.Book meeting rooms without scrolling the calendar
Book a meeting room on [date, e.g., tomorrow] from 2pm to 3pm that fits [headcount, e.g., 6 people], preferring [floor preference, e.g., 15th floor or above]. Once booked, notify [list of attendee Feishu accounts].Find that point we discussed last time — search the minutes
Search all my minutes for meetings where we discussed "[keyword, e.g., pricing strategy / tech selection / Q2 goals]". List the meeting name, date, and relevant excerpts in reverse chronological order.One meeting ID — transcript, summary, and video all handled
Take meeting ID [meeting_id] and: 1) Convert it to a minutes token; 2) Extract the AI summary and to-dos; 3) Download the recording to [local path]. Finally, post the summary to [Feishu group name].2. Messages & Groups
Let AI send messages under your own identity
Send to [Group Name]: "Something came up at 3pm today — let's push the meeting back by an hour. Sorry for the change!" Send it as me.No time to read the whole thread? Let AI summarize it
Open the full thread for message [URL or message ID] and analyze: 1) The core question; 2) Each side's main viewpoints; 3) Whether a conclusion or to-do was reached. Send me a summary under 300 words.Manage @-mentions in groups without missing anything
Scan recent unread messages in [Group Name]: 1) For every @-mention, decide whether a reply is needed and list those for me to confirm; 2) For important messages I haven't reacted to yet, add a 👀 emoji to signal "I've seen this"; 3) If the group's name hasn't been updated in 90+ days, ask if I want to update it.Send messages as you — finally an Agent that sounds like you
Send a message to [contact name / Feishu account] as me: "[Message content]". Match my usual tone: [style description, e.g., concise / polite / uses emoji].3. Docs & Content Creation
Markdown to Feishu Doc in one shot, formatting preserved
Create a Feishu Doc from this Markdown. Keep the layout clean — heading levels, code blocks, and tables must all be preserved. Title it [Title].Let AI review your draft and comment in-line
{{Doc URL}} Read this doc as if you were an external user. Check whether it's clear and concise. Don't edit the doc directly — instead, leave inline comments on passages that need improvement, with your suggestions in the comments.Explained it three times? Time to make it a doc
Scan my recent minutes and find topics I've explained 3 or more times. Turn each into a standardized doc and publish it to Feishu Cloud Docs.Search three places at once — never lose that file again
Find content related to "[keyword, e.g., quarterly review]". Search across: 1) Cloud drive and Wiki; 2) Feishu message history. Produce a list with title, source, and link, sorted by relevance.Describe the architecture, get a whiteboard diagram
On Feishu Whiteboard [URL], generate a system architecture diagram with these service nodes: [Service 1], [Service 2], [Service 3]. Connections: [Service 1] calls [Service 2] via REST API; [Service 2] notifies [Service 3] via a message queue. Use auto-layout and add icons to the nodes.Share a doc with someone — find them and grant access
Share view access for doc [URL] with: [Name 1], [Name 2]. Look up their contact info first, confirm, then add permissions, and tell me when done.Insert or download images inside docs at will
On doc [URL]: 1) Insert local image [file path] after paragraph [X]; 2) Download every image and attachment in the doc to [local folder path], keeping original filenames.When someone edits the doc, AI replies to comments for you
Monitor activity on doc [URL]: 1) When someone @[my name] in a new comment, analyze and reply; 2) When my replies receive a reaction, log it as "read"; 3) For comments unanswered for 24+ hours, remind me.Generate slide decks straight from an outline
Generate a Feishu Slides deck from this outline: title "[Deck Title]", [N] pages total. Page 1: [cover content]; Page 2: [content]; ... Send me the link when it's done.4. Email Management
Send emails with images in one step — no manual uploads
Write a follow-up email to [recipient name / email] and attach these two screenshots [local path 1] [local path 2]. Use a formal tone, subject "[XX progress sync]". Show me the draft for confirmation before sending.Auto-classify unread mail, push the important ones to a group
Go through all my unread email. For important ones (those needing my reply or decision), post a summary to [project group name] and draft a reply for each.Regret hitting send? You still have a chance
Send an email to [recipient], subject [subject], body [content]. Show me the draft first; if I say "undo" within 30 seconds of sending, recall it immediately.Never miss an email — listen, scan, and follow threads
Email handling: 1) Listen for new mail in real time; 2) On arrival, scan subject and snippet and classify as "needs my action" / "FYI" / "ignore"; 3) For "needs my action", read the full thread, generate suggested next steps, and push to [Feishu group name].Email signature attached automatically — sounds like you
Send a business email to [recipient], subject "[subject]", body [content outline]. Auto-append my signature in the format: name / role / company / contact info.5. Tasks & Planning
One sentence each morning, your day's plan auto-generated
Look at today's calendar and all my unfinished tasks. Build a prioritized list of today's focus items, marking which ones have hard deadlines.All my tasks for today, summarized in one go
Summarize today's tasks for me: 1) Query "all tasks related to me" (created by me + owned by me + assigned to me); 2) Filter for items due today or overdue; 3) Sort by priority and send me a today's plan.Tasks aren't just one line — subtasks, comments, and reminders included
On task "[Task Name]": 1) Create 3 subtasks: [Subtask 1], [Subtask 2], [Subtask 3]; 2) Add comment "[note]"; 3) Set a reminder 1 day before due. Send me the task link when done.Project tasks as a tree, with reminders set automatically
Create this task structure in Feishu Tasks: parent task "[Project]" due [date]; subtasks: [Subtask 1] owner [Name 1] due [Date 1]; [Subtask 2] owner [Name 2] due [Date 2]. Enable a 1-day-before reminder on every subtask.When a new task lands, AI responds immediately
Subscribe to Feishu task events: 1) When a new task is assigned to me, analyze it, judge priority (urgent / normal), and ping me: "New task: [name], due [date], suggested priority: [High/Med/Low]"; 2) When the status of a task I own changes, tell me what changed.6. Base & Sheets
Quick lookups in Sheets — AI locates and analyzes
In sheet [URL], find every row related to [customer name / keyword]. List their [field names, e.g., contract amount, status, contact], and compute the total and average.Base form submissions auto-trigger downstream flows
Look at the form in [Base name] and configure automation for new submissions: 1) New submission → message [owner Feishu account]; 2) Status changes to "In progress" → notify the submitter; 3) Status changes to "Done" → send a confirmation message and close the record.Manage Base attachments — let AI upload them for you
In Base [URL], in table [table name], find the record matching [field value, e.g., customer name = XXX Corp] and upload local file [file path] into the "Attachment" field.Base role permissions — different people see different data
Configure permissions in Base [URL]: 1) Create a "Sales Team" role with view-only access to the "Customers" table; 2) Create an "Ops Team" role that can edit the "Campaign Data" table; 3) Assign these people to roles: [name → role mapping].7. Cloud Drive & Wiki
One click to organize chat images into the Drive
Scan messages in [Group Name] from the last 7 days, download every image and file, and upload them to Drive [folder URL] like this: images go into the "Images" folder, docs into "Documents". Confirm with me when done.Big files no longer block you — Drive cleanup in one go
Clean up Drive [folder URL]: 1) Upload the large local file [file path] (auto-chunked); 2) Export every .pptx in the folder to PDF and download to [local path]; 3) Move files from [old folder URL] into this folder.Treat the Wiki like a database — read and write nodes any time
Manage Wiki [Wiki URL]: 1) List the hierarchy of all top-level and child nodes; 2) Move doc [doc URL] under "[target node name]"; 3) Create a new child doc under "[node name]" titled "[title]".8. Automation & Approvals
Approvals piling up? AI handles the obvious ones
Query all my pending approvals and: 1) Auto-approve expense requests under [threshold, e.g., 1000 RMB]; 2) For everything else, list applicant, reason, and amount and wait for my confirmation; 3) Flag any item pending more than 3 days.Event subscriptions — AI as your 24/7 Feishu duty officer
Subscribe to these Feishu events and handle them by rule: 1) New @-mention to me in [Group Name] → analyze whether immediate reply is needed and ping me if so; 2) New approval request → if it's an expense and amount < 500, auto-approve; 3) New task assigned to me → auto-add it to today's plan.Approval stuck? AI tracks down who's holding it up
Check the latest status of my [approval type, e.g., expense / leave] request. If it's been pending for [days, e.g., 2] days, find the current approver and send them a friendly nudge — keep the tone light, not pushy.9. AI Memory & Skills
AI scans your Feishu activity and recommends Skills
Look at my Feishu activity over the last week (chats, docs, calendar, minutes) and recommend Skills that suit me, or wrap things I do repeatedly into a Skill.AI gets to know you better — self-evolving from your Feishu activity
Scan my Feishu activity from the last week (minutes, chats, docs, calendar) and update your memory of me — collaboration network, work focus, decision preferences — and write it into your memory file.Package this as a Skill — call it with one line later
Create a Feishu Skill named "[Skill name, e.g., Weekly Report Generator]" with description: [function spec, e.g., takes this week's work log and outputs a structured weekly report covering: completed items / next-week plan / blockers]. Send me the Skill so I can reference it directly.10. Advanced & Developer
No API docs? Use the Explorer to find and call them live
Use the OpenAPI Explorer to find Feishu endpoints related to "[capability keyword, e.g., meeting room / mail / approval]". List the relevant endpoint names and descriptions, then call [endpoint name] with parameters [params] and show me the response.Return only the fields you want — jq filter in one step
Query [Feishu resource, e.g., group list / calendar events / task list], return only the [field names, e.g., name/id/owner] fields, and present the results as a table.When the CLI ships a new version, AI tells you what changed
Check whether lark-cli has a new version. If yes: 1) Tell me my current version and the latest; 2) List the major updates (new commands / improvements); 3) Highlight any updates relevant to features I use often.This article is sourced from the Feishu knowledge base. See theoriginal document for full details.