Skip to content

Multi-agent Git Collaboration Rules

Multi-agent Git collaboration rules are a set of operational protocols designed to prevent concurrent agents from overwriting each other's work or causing repository instability when working within a shared codebase^[001-TODO__Pi_Monorepo_-_AI_Agent_开发工具包.md].

When multiple autonomous agents operate on the same repository simultaneously, standard "commit all" commands pose a significant risk, as an agent may indiscriminately stage and commit changes made by another active agent^[001-TODO__Pi_Monorepo_-AI_Agent_开发工具包.md]. These rules enforce strict boundaries regarding file ownership and staging behavior to ensure parallel integrity^[001-TODO__Pi_Monorepo-_AI_Agent_开发工具包.md]。

Core Principles

The fundamental goal of these rules is to establish clear lines of ownership and minimize merge conflicts^[001-TODO__Pi_Monorepo_-_AI_Agent_开发工具包.md]。

File Ownership

Agents must only interact with files they have explicitly created or modified^[001-TODO__Pi_Monorepo_-_AI_Agent_开发工具包.md]。 This creates a sandbox environment where an agent's workspace is isolated from the changes of others.

Selective Staging

The most critical rule is the prohibition of bulk staging commands^[001-TODO__Pi_Monorepo_-_AI_Agent_开发工具包.md]。

Execution Rules

The following restrictions must be enforced on the Git operations available to the agents:

  1. Restricted git add:

    • Agents MUST explicitly list specific files when staging (e.g., git add path/to/file.ts).
    • Agents MUST NOT use recursive or wildcard flags that stage all changes, such as git add -A, git add ., or git add -u^[001-TODO__Pi_Monorepo_-_AI_Agent_开发工具包.md]。

  2. Scope of Modification:

    • Agents should only modify files within their designated task scope or newly created files^[001-TODO__Pi_Monorepo_-_AI_Agent_开发工具包.md]。

Applications

This protocol is utilized in development environments where agentic workflows are parallelized to increase throughput, such as the Pi Monorepo architecture^[001-TODO__Pi_Monorepo_-AI_Agent_开发工具包.md]。 By adhering to these rules, a system can safely run multiple coding agents (e.g., one for UI, one for backend logic) against the same Git repository without constant merge conflicts or data loss^[001-TODO__Pi_Monorepo-_AI_Agent_开发工具包.md]。

  • Pi Monorepo: The AI agent development toolkit where these specific collaboration rules are implemented and enforced^[001-TODO__Pi_Monorepo_-_AI_Agent_开发工具包.md]。
  • [[Agent Runtime]]: The underlying system that manages the execution flow of agents.

Sources

  • 001-TODO__Pi_Monorepo_-_AI_Agent_开发工具包.md