GitAgent

GitAgent is a GitHub-native, model-agnostic framework designed to facilitate AI-assisted software development. It allows various AI agents such as Claude Code, Gemini, and GitHub Copilot to seamlessly manage project development by leveraging GitHub's native features as the primary source of truth.

Problem Solving Capabilities

GitAgent addresses several challenges in AI-assisted development:

Key Features

• Model Agnostic: Compatible with any AI agent capable of reading files and utilizing the GitHub CLI.
• GitHub Native: Utilizes GitHub's Issues, Projects, Labels, and Actions for effective state management.
• Human Visibility: Facilitates clear tracking of project status without the need to sift through extensive AI-generated outputs.
• Gate System: Requires human approval at key phases using commands like /approve and /verify-pass.
• Flexible Workflows: Allows for various methodologies, including Test-Driven Development (TDD), flexible implementation, or spec-first approaches.
• Context Engineering: Ensures that AI agents maintain focus through structured context management.

Usage

To begin using GitAgent, follow these steps:

1. Create a new repository using the GitAgent template:

   gh repo create my-project --template MauricioPerera/gitagent --public --clone
   cd my-project
   

2. Run the setup workflow to initialize project labels and issues:

   gh workflow run setup.yml
   

3. Configure the project by editing files in the .gitagent/context/ directory. Here is a snippet for defining the product vision:

   ## Product Vision
   Build a task management app for remote teams...
   

4. Engage with the AI by passing commands like:

   claude
   > Read AGENT.md and help me build a user authentication feature
   

Workflow Overview

GitAgent enables an organized development cycle facilitated by AI and human collaboration. Below are the phases of interaction:

1. A user creates a tracking issue, initiating the workflow.
2. The AI agent reads the AGENT.md protocol and context files, generating a specification.
3. The specification is submitted for human review and requires approval.
4. The AI generates an implementation plan, which must again be approved by a human.
5. After receiving approval, the AI implements the changes and commits code, marking phases with appropriate labels.
6. Finally, once all tasks are completed, the project status is updated to "done" and summaries are provided.

Commands

Human engagement is crucial in this process and is facilitated through simple commands:

Project Structure

GitAgent's structure is organized for easy navigation and understanding, featuring templates and configuration files necessary for effective collaboration:

your-project/
├── AGENT.md  # AI agent protocol entry point
├── .gitagent/
├── docs/     # Additional documentation
└── specs/    # Generated specifications and plans

GitAgent streamlines the integration of AI in development while ensuring human oversight and approval, making it an invaluable tool in modern software development practices.