larksuite/cli

lark-cli – A Command‑Line Interface for Lark/Feishu

What it is – An open‑source CLI (written in Go/Node) that lets humans and AI agents talk to the Lark/Feishu productivity suite. It wraps more than 2 000 Open Platform APIs behind 200+ ready‑to‑use commands and a set of “skills” that are deliberately structured for AI‑driven automation.

Key capabilities

• Broad coverage – Calendar, messenger, docs, drive, sheets, slides, tasks, mail, meetings, OKR, wiki, contacts, approvals, attendance, Base tables, etc.
• Three‑layer command model
  1. Shortcuts (+agenda, +messages‑send) – human‑ and AI‑friendly, with defaults and table output.
  2. API commands – auto‑generated one‑to‑one mappings to Lark Open API endpoints.
  3. Raw API – direct calls to any of the 2 500+ endpoints.
• Agent‑native “skills” – 24 pre‑packaged skill modules (e.g., lark‑calendar, lark‑im, lark‑doc) that expose a clean, typed interface for AI agents, making it easy to embed the CLI in bots or other automation pipelines.
• Security‑first – OS key‑chain credential storage, input‑injection protection, optional DPoP proof‑of‑possession, and a built‑in risk‑control signal that reports only minimal device info to Feishu.
• Installation – One‑liner via npx @larksuite/cli@latest install or build from source with Go 1.23+.
• Authentication – OAuth flow with interactive TUI, device‑code mode for agents, and fine‑grained scope selection.

Who should use it

• Individual developers who want a quick terminal interface to Lark services.
• Teams building AI assistants, bots, or RPA pipelines that need reliable, structured access to Lark data.
• Enterprises that want to embed the CLI in internal tools, with centralized credential stores and audit logging.

How it works (high‑level)

1. Install the package.
2. Run lark-cli config init to register an app and store its client ID/secret.
3. Authenticate (lark-cli auth login).
4. Issue commands – either a shortcut (lark-cli calendar +agenda) or a raw API call (lark-cli api GET /open-apis/calendar/v4/calendars).

Output formats – JSON (default), pretty‑printed, table, CSV, NDJSON, making it easy to pipe results into other scripts or AI pipelines.

Risks & mitigations – The README warns that AI agents can unintentionally perform harmful actions; default protections (input sanitisation, limited token‑use signals) are enabled, and users are advised not to relax them.

License – MIT, with the usual requirement to obey Lark/Feishu service terms.

---

All details are taken directly from the repository’s README.