What problem does it solve? It lets an AI agent safely operate a Feishu (Lark) mailbox on the user's behalf—reading, searching, drafting, replying, forwarding, and organizing email—while enforcing strict confirmation rules so no message is ever sent without explicit user approval. ## Core Features & Use Cases - Full mail lifecycle: Triage the inbox, read single or batched messages and threads, create and edit drafts, reply, reply-all, forward, schedule sends, and cancel scheduled sends through lark-cli mail shortcuts. - Organization and automation: Modify labels, read states, and folders, soft-delete messages, manage inbox rules, signatures, templates, and read receipts. - Safety-first design: Treats email content as untrusted input against prompt injection, requires user confirmation before any send or destructive action, and defaults all compose operations to drafts. - Use Case: Ask the agent to summarize today's unread inbox, draft an HTML reply to a client with an attachment, show you the draft link, and send it only after you confirm. ## Quick Start Ask the agent to check my Feishu inbox and draft a reply to the latest email from my manager.