Lark doc skill: read, write and edit Feishu docs with an agent

Source: larksuite/cli · Maintainer: Lark · License: MIT · Checked: 26 October 2026

What is the lark-doc skill?

The lark-doc skill lets an AI coding agent such as Claude Code or Codex read, write and edit Lark and Feishu cloud documents through the official lark-cli command line tool. It is for teams that keep specs, meeting notes and reports in Lark Docs and want an agent to work on them directly.

It lives in larksuite/cli, which the README calls "The official Lark/Feishu CLI tool, maintained by the larksuite team" and says is "built for humans and AI Agents." The README's skill table sums lark-doc up as "Create, read, update, search documents (Markdown-based)".

Language note: the SKILL.md and its reference files are written in Simplified Chinese. Commands, flags and file names are in English. This page describes the skill in English; the quotes below come from the English README, and the description below is Kipt's translation, not a quote.

Translated, the skill's description says it handles Feishu cloud document (Docx and Wiki) content: reading, creating and editing documents, inserting or downloading images and attachments, and working with mind notes. It triggers when you give a document URL or token, including doubao.com /docx/ and /wiki/ links, and it routes by URL path and token rather than domain.

As of 26 October 2026, its skills.sh page listed about 478,900 installs (snapshot, 26 October 2026).

What does it do?

The SKILL.md is short. It works as a router: it tells the agent to decide the scenario first, then read only the reference file for that stage, once. The core commands are lark-cli "shortcuts" (commands prefixed with +):

• Read and summarize: docs +fetch. Reads a whole document or just part of it. The fetch reference lists scopes: outline for the heading list, section for one heading's section, range between two block IDs, and keyword for matches. A --detail flag picks simple, with-ids or full.
• Write from scratch: the create workflow. An eight-step authoring process. The agent works out the reader and their task, picks a genre "contract", gathers material, records a presentation decision as JSON, drafts XML, checks it with docs +script --command parse, then creates the doc.
• Import or blank doc: docs +create. Skips the workflow when you only want an empty document or to import content you supplied as is.
• Edit: docs +update. Commands include str_replace, block_replace, block_insert_after, block_delete and block_move_after. The update reference asks the agent to fetch, diagnose, plan a minimal patch, apply it, then fetch again to verify.
• History: +history-list and +history-revert. Look up and roll back document versions.
• Media, covers, whiteboards and mind notes. Insert local images or files, preview or download media, manage a Docx cover, and hand whiteboard work to the lark-whiteboard skill.

The references/genres/ folder holds 28 files. Ten are routing files (nine content profiles such as workplace, report and knowledge, plus one adapter for email, WeChat and Xiaohongshu) and 18 are genre guides, such as PRD, meeting minutes, weekly report, white paper, technical doc, SOP or tutorial, email, WeChat article and Xiaohongshu post.

Two details matter in practice. Documents default to Lark's XML format (DocxXML); Markdown is used only when you ask for it or need Markdown fidelity, even though the README calls the skill "Markdown-based". And the skill sends some work elsewhere: file-level Drive actions and standalone comments go to lark-drive, and data inside sheets or Base tables is out of scope.

How do you install it?

You need three things: the lark-cli binary, the lark-shared skill (the SKILL.md lists both under requires), and a signed-in Lark or Feishu account.

Official route (CLI plus all skills). From the README:

npx @larksuite/cli@latest install
lark-cli config init
lark-cli auth login --recommend

The install wizard in the repo then installs the skills globally, first from an open.feishu.cn source and, if that fails, from larksuite/cli. That is why skills.sh also shows a separate open.feishu.cn entry for lark-doc.

Just this skill, with its dependency. The command on skills.sh is npx skills add https://github.com/larksuite/cli --skill lark-doc. The skills CLI takes --skill more than once, so add lark-shared in the same step:

npx skills add https://github.com/larksuite/cli --skill lark-doc --skill lark-shared

Add -g to install for all projects. In Claude Code the folders land in .claude/skills/, or ~/.claude/skills/ with -g, as the Claude Code skills docs describe. Codex and Cursor read the same SKILL.md format; see Codex skills and Cursor skills.

Installing the skill alone does not install lark-cli. Without the binary and a login, the agent has nothing to call.

How do you use it?

Paste a document link or token and say what you want. The skill recommends running document commands as your user, with --as user, so the agent sees the same documents you do.

Example prompts (suggestions, not quotes from the repo):

• Summarize this Lark doc and list the open decisions: <doc URL>
• Find the section about the release plan in <doc URL> and rewrite it as a numbered checklist. Don't touch other sections.
• Write a one-page PRD for the new export feature in Lark Docs, using the notes in ./notes.md.
• Turn today's meeting notes in <wiki URL> into meeting minutes with owners and due dates.
• Roll this doc back to the version from yesterday morning: <doc URL>

For edits, expect the agent to fetch a small part of the document with block IDs first, then patch only those blocks. The update reference warns against overwrite unless you ask for a full rebuild, because it can lose comments and unsupported resources.

Tips and limits

• Audits (12 October 2026). On skills.sh, Gen Agent Trust Hub rated lark-doc Pass and Socket rated it Pass. Snyk rated it Warn, with one medium finding (W011): third-party content exposure, because docs +fetch reads documents and comments that other people may have written. Gen Agent Trust Hub noted the same indirect prompt injection surface. See prompt injection in coding agents for the general risk.
• The agent acts as you. The README's security section says: "After you authorize Lark/Feishu permissions, the AI Agent will act under your user identity within the authorized scope, which may lead to high-risk consequences such as leakage of sensitive data or unauthorized operations." Grant only the scopes you need.
• Confirmation gates. The lark-shared skill tells the agent to confirm your intent before write or delete operations, and treats exit code 10 as a high-risk confirmation gate that needs your explicit approval.
• Keep files in the working directory. lark-doc prefers @./relative/path references for local files. The shared rules in lark-shared go further: path arguments such as --file and --output accept only relative paths under the current directory, and absolute paths fail with unsafe file path.
• Chinese instructions. If you want to read or edit the skill yourself, you will need Chinese or a translation.
• Separate terms apply. The code is MIT, but the README notes that using the Lark and Feishu APIs means following their user terms and privacy policies.

Save it in your Kipt library

The repo's MIT License lets you copy and change the skill as long as you keep the copyright notice. In Kipt, open Skills and paste the folder link https://github.com/larksuite/cli/tree/main/skills/lark-doc into the import box; Kipt reads the SKILL.md and its reference files. The repo has many skills, so importing the repo link shows a list where you can also pick lark-shared. Your copy lives in your own private GitHub repo and installs into Claude Code or Codex with one command:

npx skills add https://github.com/<you>/<repo>.git --all -g

Kipt is not affiliated with Lark, Feishu or larksuite.

Similar skills

• Doc co-authoring skill: Anthropic's skill for drafting documents with you section by section.
• DOCX skill: create and edit Word files locally instead of Lark cloud docs.
• Internal comms skill: status updates and newsletters in a company's house style.
• Claude Code skills: how skills load and where they live.