Multi-Bot Collaboration: A Shared Git Workspace for Unified AI Agent Coordination
The article presents a practical framework for coordinating multiple AI bots using a shared Git repository as a centralized workspace, detailing directory structure, core principles like single source of truth and content-based organization, and how this avoids bottlenecks such as rule conflicts, context limits, and fragmented memories.
Introduction
The author has followed personal agent products like OpenClaw since early days but found them far from production-grade compared to tools like Claude Code or Codex. With product iteration and stronger models, they explored turning such agents into productivity tools. This article uses Grok Bot as a case study to demonstrate a shared workspace approach.
Bottlenecks of AI Bots
After extended use, several limitations emerged:
Each bot's internal model is opaque; cannot customize model selection per task difficulty.
Rules accumulate per bot, causing role confusion and regression of corrected behaviors.
Limited context for complex codebases; bots suit only simple or low-context tasks.
AI Bot + cloud computer mode risks losing important files; only suitable for temporary files.
In group chats, multiple bots talk past each other; unclear responsibilities and insufficient model capability hinder progress.
The author started with a single "Chief of Staff" bot, but rule overload made it lose focus. Splitting into multiple bots and adding them to a group introduced coordination problems. The insight: splitting is necessary, but true coordination comes from accumulating experience in the right place — a shared project with one repository, one entry point, one division-of-labor document. Each bot keeps independent conversation, memory, and description, but they share the same computer, files, and repository.
Project-Centered Workspace Design
A local workspace centered on "Team Bots" adopts a coding-agent development mindset with proper project rules, enabling all bots to collaborate and iterate continuously. The design is generalized to avoid vendor lock-in when switching personal agent products.
Directory Structure
personal-workspace/
├── AGENTS.md # Entry point for all bots, inspired by Coding Agent
├── README.md # Directory overview and index
├── 00 Todo/ # Daily TODO tasks, maintained by manager bot
│ └── 2026-10 TODO.md
├── Personal/ # User profile, habits, and bot role definitions
│ ├── 个人介绍.md
│ ├── 工作习惯.md
│ └── Bot 分工.md
├── Knowledge/ # Knowledge base for multi-platform content prep and distribution logs
│ ├── README.md
│ ├── wechat-index.md # Published article index, real-time updated to avoid topic conflicts
│ └── 00 AI/示例系列/
│ ├── 01-示例文章.md
│ ├── meta.yaml
│ └── publish/
├── Media/
│ └── README.md
├── Company/ # Company/studio public info, project overviews
│ ├── 00 概况.md
│ └── 01 项目/
│ └── 示例项目.md # Build, release rules and logs
└── Private/ # Default unread, excluded from GitCore Design Principles
Workspace is a Git repository; all bots read rules, write outputs, and commit within it.
Every bot's rules live in the workspace; everything is visible in the workspace to avoid hidden rules.
Directories are organized by content, not by bot; a bot can read multiple directories, and a directory can be read by multiple bots.
Single source of truth: each piece of data stored once, referenced elsewhere to simplify maintenance.
Root AGENTS.md only directs bots where to read, not containing rules itself. It has two phases: first read personal intro, habits, bot roles, directory overview; then read task-specific files (e.g., Knowledge/README.md for writing, project docs for releases).
Conclusion
Using AI bots for chat and task delegation is easy. The hard part is making them follow one rule set, produce consistently, and handle tasks from simple to complex to genuinely free the user. The shared Git workspace achieves this by centralizing rules, context, and outputs while keeping bot identities separate.
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
