zarazhangrui/lark-coding-agent-bridge

lark-channel-bridge lets you talk to AI coding assistants (Claude Code or Codex CLI) directly from Feishu or Lark, the popular workplace messaging apps. Instead of switching to your terminal every time you want to ask your coding agent something, you just send a message in chat. You can DM the bot privately or mention it in group conversations, and it forwards your messages to whichever coding tool you've set up on your computer. The setup is straightforward: you install it globally with npm, run one command, and scan a QR code with your Feishu/Lark app to connect. The bridge then sits between your chat and your local coding agent. When you send a message, it passes it along and streams the response back to you in real time on a chat card. Each conversation keeps its own context, so you can have multiple ongoing discussions without them getting confused. If you send several messages quickly, it batches them together, and commands like /new or /stop can interrupt what the bot is currently doing. This is useful for developers or teams who already use Feishu/Lark as their communication tool and want quick access to a coding assistant without leaving their chat workflow. For example, a solo developer could ask the bot to help debug a snippet from their phone during a commute, or a team could share a bot in a project group chat where anyone can ask questions. You can switch between project directories using /cd, save common ones as named workspaces with /ws save, and even send images or files directly in chat for the agent to work with. Access control is private by default: only the person who set up the bot can use it, and everyone else's messages are silently ignored. You can selectively invite specific people, open up particular group chats, or designate admins who get broader access. The bot also supports running separate profiles for Claude and Codex simultaneously, so you could have one bot for each tool if needed. One notable tradeoff is that the working directory you set isn't a security sandbox, it just sets where the agent starts looking for files. Actual file access depends on the coding agent's own permission settings, which you can configure to range from full access to read-only mode.