ClawBro

<div align="center"> <img src="./assets/logo.png" alt="clawBro Logo" width="200"> <h1>🦀 clawBro: Let CLI Coding Agents Work Like OpenClaw in Chat and Team Mode</h1> <p> <strong>Built around OpenClaw ideas, clawBro helps Claude Code, Codex, Qwen, Qoder, Gemini, and other CLI Coding Agents work together across WeChat, Lark, DingTalk, and long-running team workflows.</strong> </p> <p> <a href="./README_ZH.md"><strong>中文</strong></a> · <a href="./README_JA.md"><strong>日本語</strong></a> · <a href="./README_KO.md"><strong>한국어</strong></a> </p> <p> <a href="#-project-status">Project Status</a> · <a href="#️-architecture">Architecture</a> · <a href="#-use-cases">Use Cases</a> · <a href="#-quick-start">Quick Start</a> · <a href="#-team-modes">Team Modes</a> · <a href="#-cli-coding-agent-integration">CLI Coding Agent Integration</a> · <a href="./docs/setup.md">Setup Guide</a> </p> <p> <img src="https://img.shields.io/badge/version-0.1.10-blue" alt="Version"> <img src="https://img.shields.io/badge/rust-1.90%2B-orange" alt="Rust"> <img src="https://img.shields.io/badge/license-MIT-green" alt="License"> <img src="https://img.shields.io/badge/agents-Claude%20%7C%20Codex%20%7C%20Qwen%20%7C%20Qoder%20%7C%20Gemini-111827" alt="Agents"> <img src="https://img.shields.io/badge/channels-WeChat%20%7C%20Lark%20%7C%20DingTalk-4EA1FF" alt="Channels"> <img src="https://img.shields.io/badge/runtime-Native%20%7C%20CLI%20Bridge%20%7C%20OpenClaw-8B5CF6" alt="Runtime"> <img src="https://img.shields.io/badge/modes-Solo%20%7C%20Multi%20%7C%20Team-111827" alt="Modes"> </p> </div>

clawBro is a Rust-based system for making CLI Coding Agents work together across local workflows, chat apps, and long-running team collaboration.

It stays close to the OpenClaw spirit, but pushes toward practical teamwork: Claude Code, Codex, Qwen, Qoder, Gemini, and related coding agents can be organized into solo, role-based, and lead-plus-specialist workflows, then connected to WeChat, Lark, DingTalk, and WebSocket entrypoints.

It also supports the official WeChat lobster path: a WeChat front bot can lead shrimp soldiers and crab generals in Team mode, keep the user-facing conversation stable, and delegate work to specialists behind the scenes.

📢 Project Status

• [03-20] Durable scheduling is now built in: one-shot reminders, exact-time jobs, interval polling, cron schedules, chat-created reminders, and current-session cleanup all share the same runtime scheduler.
• [03-19] One clawbro surface now brings together multiple AI CLI Coding Agents instead of forcing one tool per workflow.
• [03-19] Team orchestration supports lead-driven workflows, specialist agents, milestone delivery, and named roles like planner, coder, reviewer, and researcher.
• [03-19] Group and direct-message usage now fit the same routing model, with WeChat DM, Lark, DingTalk Stream Mode, DingTalk custom robot webhook, and WebSocket entrypoints.
• [03-19] Multi-IM connectivity is now practical for always-on chat workflows: one runtime can stay online across WeChat, Lark, DingTalk, and team conversations at the same time.
• [03-19] Operational controls include approvals, allowlists, memory-aware sessions, /health, /status, /doctor, and diagnostics surfaces.

clawBro is built for engineering, research, and workflow experimentation. It is meant for real agent collaboration, not just another chat wrapper.

📸 Demo

<table align="center"> <tr align="center"> <td><img src="./assets/demo-multi-bot-mention.png" width="260"><br>_{<b>Mention multiple bots at once in group chat</b>}</td> <td><img src="./assets/demo-scheduled-tasks.png" width="260"><br>_{<b>Scheduled task reminders via chat</b>}</td> <td><img src="./assets/demo-wechat-team.png" width="260"><br>_{<b>Team mode collaboration in WeChat</b>}</td> </tr> </table>

Key Features of clawBro:

🏛️ Unified Control Plane: One clawbro entrypoint for setup, routing, session management, diagnostics, and runtime dispatch.

🤖 Unified CLI Coding Agents: Bring Claude, Codex, Qwen, Qoder, Gemini, and other coding CLIs into one product surface instead of juggling separate entrypoints.

👥 Team Orchestration: Support solo, multi, and team interaction models with lead + specialists, scope-aware routing, and milestone-style collaboration.

💬 Group Chat Collaboration: Connect workflows to Lark and DingTalk, route group mentions to named agents, and turn chat rooms into AI workbenches.

📡 Always-On Multi-IM: Keep one clawbro runtime online across Lark, DingTalk Stream Mode, DingTalk custom robot webhook, and WebSocket, then keep chatting without switching tools.

🧠 Memory and Habits: Let agents accumulate working memory, repeated preferences, review standards, and recurring project context over time.

⏰ Durable Scheduling: Create reminders, one-time jobs, recurring polling loops, and agent-driven scheduled work through delay, at, every, and cron.

🛡️ Operationally Controllable: Built-in config validation, approval flow, allowlists, doctor/status commands, and health endpoints.

🏗️ Architecture

User / Group / WebSocket / Scheduled Jobs
              |
              v
           clawbro
              |
              +--> Routing / Session / Memory / Bindings / Team
              |
              +--> ClawBro Native ------> runtime-bridge ------> clawbro-agent-sdk
              |
              +--> Coding CLI Bridge ---> Claude / Codex / Qwen / Qoder / Gemini / custom coding CLIs
              |
              +--> OpenClaw Gateway ----> remote agent runtime
              |
              +--> Channels ------------> Lark / DingTalk / WebSocket delivery

Table of Contents

• Project Status
• Key Features
• Architecture
• Features
• Use Cases
• Install
• Quick Start
• Scheduled Tasks
• Team Modes
• CLI Coding Agent Integration
• Chat Channels
• Configuration & Operations
• Project Structure
• Documentation Map
• Positioning

✨ Features

<table align="center"> <tr align="center"> <th><p align="center">🤖 Coding Agent Hub</p></th> <th><p align="center">👥 Team Coordination</p></th> <th><p align="center">🧠 Memory Habits</p></th> </tr> <tr> <td align="center">One control plane for Claude, Codex, Qwen, Qoder, Gemini, and other coding-focused agents.</td> <td align="center">Lead + specialists, scope-aware team routing, group mentions, milestone delivery, and task-oriented collaboration.</td> <td align="center">Shared memory and agent memory help the system keep long-running habits, context, and preferences.</td> </tr> </table>

🌟 Use Cases

🚀 Full-Stack App Building

Turn one request into a coordinated build loop:

• @planner breaks the product request into milestones
• @coder implements API routes, UI flows, and data models
• @reviewer checks quality, risks, and regressions
• @tester fills in edge cases and missing validation

In Team mode, the lead can keep the user-facing conversation clean while specialists work in the background. In group chat, the same setup can feel like an AI project room instead of a single bot window.

📚 Deep Research and Report Writing

Use ClawBro as a research squad:

• @researcher collects source material
• @critic looks for gaps, counterexamples, and weak assumptions
• @writer turns the findings into a structured report
• the lead agent summarizes progress and final conclusions

This works especially well for technical reports, architecture comparisons, literature reviews, and long-form analysis that benefits from multiple perspectives before one final answer.

🧑‍💻 PR Review and Design Review

Drop a patch, PR, or design note into a chat and route it to the right mix of agents:

• @coder focuses on implementation details
• @reviewer checks correctness and maintainability
• @researcher verifies outside dependencies or competing approaches
• the lead returns a consolidated recommendation

This gives you something closer to an AI review room than a single one-shot answer.

💬 Group Chat With Multiple Named Agents

ClawBro is a natural fit for role-based group workflows:

• @planner for decomposition
• @coder for implementation
• @reviewer for criticism
• @researcher for evidence gathering

That pattern works for engineering teams, study groups, product discussions, and internal support rooms. Even when the strongest current Team path is still lead-driven, the group experience can already feel much more structured than a generic bot chat.

🧠 Memory-Driven Coding Habits

ClawBro is not just about one conversation at a time. Over repeated use, it can preserve working context such as:

• architecture preferences
• recurring review standards
• naming conventions
• project-specific workflows
• things a user repeatedly asks the system to remember

That makes it useful for building a long-running coding habit, where your agents gradually become more aligned with how you actually work instead of resetting to zero every day.

🎭 Fun Team Play: Werewolf, RPG, and Role Rooms

The same role system also works for playful group scenarios:

• a lead agent can act as the moderator in Werewolf
• specialist agents can play judge, narrator, analyst, or character roles
• role-based group chats can simulate product debates, mock trial rooms, or scripted multi-character conversations

This is one of the most distinctive parts of the project: the architecture is serious enough for engineering work, but flexible