Lark / Feishu CLI

Pick a task and get started

Install and configure your account in a personal terminal, then give the assistant a document URL. Start with document-read access and add only the permissions required by later Base tasks.

1. Install the CLI and companion skills in two steps

   Copy install command

   These two steps install the reviewed CLI release 1.0.98 and the current repository skills globally. The author’s one-step npx @larksuite/cli@latest install also proceeds to app configuration and authorization in an interactive terminal. Choose Feishu or Lark before configuring, then check the installed command help.

   npm install -g @larksuite/[email protected]
   npx skills add larksuite/cli -y -g

   Read the author’s setup guide

   First run: choose the platform and fetch an outline

   Create brief.md in the current workspace with a full Docx/Wiki URL you can read, the relevant sections, and the summary goal. Run the two installation commands above. On a new setup, run lark-cli config init and choose an existing or new app and Feishu or Lark in the interactive form. Reuse an existing configuration instead of initializing it again. Complete the platform confirmation yourself, then run lark-cli auth login --scope "docx:document:readonly". Check brand, identity, verified, and the user status with lark-cli auth status --json --verify; verification contacts the server. Paste task one into the assistant and check that its headings and block IDs belong to your document. If the app has not enabled the scope, its administrator must enable it before user authorization.

   Read the identity and authorization guide

2. Give your AI this first task

   Copy prompt

   These are editorial operating examples. Replace URLs, paths, fields, and IDs with your own. The outputs are expected deliverables, not results from our use of an account.

   Use lark-doc and lark-cli to read the document in brief.md and summarize decisions, actions, and open questions. Explicitly use --as user throughout. Read the online document and write a local summary without changing its content. Check lark-cli docs +fetch --help. Fetch the outline with lark-cli docs +fetch --doc "REAL_DOCUMENT_URL" --scope outline --max-depth 3 --doc-format xml --detail with-ids --as user --format json. Use returned headings and block IDs to select sections, then fetch with --scope section --start-block-id "REAL_BLOCK_ID", keeping XML and with-ids. Save complete JSON responses under source/ in the current workspace, retaining content, reference_map, revision_id, and tips within data.document together. State the coverage of excerpts, truncated comments, and embedded-table references. Retrieve missing relevant content before summarizing it; do not present fragments as a full read. Compare revision_id across the outline and section responses. Where the API supports that historical revision, pass --revision-id on subsequent reads to keep one version. If the document has changed and the earlier revision cannot be fetched, refresh the outline and relevant sections and state the version summarized. Write summary.md with source headings and block IDs for each decision or action. Use only stated owners and deadlines, marking missing details. Keep document statements, comment opinions, and your own synthesis distinct. Check the business response ok, identity, and exit status, then spot-check the summary against source. For permission failures, identify the missing scope or resource access and retain the user identity rather than switching to bot automatically.

3. Check what you receive

   • A decision/action summary with document revision, headings, and block IDs.
   • Complete source JSON, preserving document and comment references together.
   • Coverage limits and a record of source spot-checks.
   Read the author’s notes for this task