Feishu CLI Gives AI Agents Hands to Operate Your Entire Office Suite
Most AI-in-the-workplace demos stop at generating text; this integration closes the loop by letting the agent execute the final step inside the tools teams already use. For any developer or team running on Feishu/Lark, it eliminates the manual copy-paste tax on routine ops—spreadsheet formatting, report assembly, meeting scheduling—and shifts the bottleneck from tool operation to clear thinking.
The Feishu CLI (lark-cli) bridges the gap between AI chat and actual work by giving language models direct control over Feishu's office tools. Instead of copying AI-generated text into documents or manually updating spreadsheets, a developer can install the CLI inside an agent like Codex and issue plain-English commands that create cloud docs, populate multidimensional tables, configure automations, generate slide decks, draw vector diagrams on whiteboards, and send personalized messages to colleagues.
A walkthrough covers eight practical scenarios: drafting and reviewing documents with inline comments, building project trackers with automated deadline reminders, cross-table aggregation for weekly reports, generating onboarding slide decks and fashion ordering fair presentations, converting Mermaid diagrams into editable whiteboard flowcharts, running JS-driven high-density infographics like radar charts and product roadmaps, batch-sending individualized notifications, and extracting action items from meeting minutes into Feishu Tasks.
The tool covers 100+ skills under an MIT license and requires only Node.js and a few CLI commands to set up. A restart of the agent host is mandatory for skills to load, and OAuth authorization determines whether the AI can read personal data like calendars and private messages.
The real unlock is not the CLI itself but the combination of a local agent runtime (Codex) with a permissioned API surface—this turns the AI from a text generator into an authenticated user that can act on your behalf across an entire office suite.
Requiring a full restart of Codex for skills to load is a sharp edge that will trip up first-time users; it signals that the agent-plugin architecture is still early and not designed for seamless discovery.
The whiteboard's ability to run JavaScript for coordinate-level layout pushes it past simple diagram generation into a programmable visualization engine, which is unusual for an office suite integration.
Cross-table aggregation and automated dashboard creation hint at a future where weekly reporting is a fully automated pipeline, not a recurring human task—provided the underlying tables are kept current.
The authorization model creates a meaningful split: without OAuth, the AI is a public-space operator; with it, the AI becomes a full delegate with access to private calendar and message data, which raises trust and security questions that the guide does not address.