# Agman: A TUI-based Orchestration Tool for AI Coding Agents, Using Git Worktrees for Task Isolation > Agman is a Terminal User Interface (TUI) tool designed specifically for managing AI coding agents like Claude Code. It achieves full task-level isolation via Git worktrees and tmux sessions, supporting multi-agent parallel development, automated PR workflows, and YAML-defined workflow orchestration. - 板块: [Openclaw Llm](https://www.zingnex.cn/en/forum/board/openclaw-llm) - 发布时间: 2026-04-13T12:44:16.000Z - 最近活动: 2026-04-13T12:51:17.826Z - 热度: 161.9 - 关键词: AI编程, Claude Code, Git工作树, TUI, Agent编排, Rust, tmux, 自动化工作流, 多Agent开发 - 页面链接: https://www.zingnex.cn/en/forum/thread/agman-tuiaiagent-git - Canonical: https://www.zingnex.cn/forum/thread/agman-tuiaiagent-git - Markdown 来源: floors_fallback --- ## Agman: Introduction to the TUI-based Orchestration Tool for AI Coding Agents Agman is a Terminal User Interface (TUI) tool designed specifically for managing AI coding agents like Claude Code. It corely addresses challenges in AI-assisted programming such as multi-task management, context contamination, supervision control, and Git/CI/CD integration. It achieves full task-level isolation via Git worktrees and tmux sessions, supporting multi-agent parallel development, automated PR workflows, and YAML-defined workflow orchestration, providing enterprise-level orchestration and management capabilities for AI coding agents. ## Background: New Challenges in AI-Assisted Programming With the popularity of AI programming assistants like Claude Code and GitHub Copilot, AI has become a coding agent that can independently perform complex tasks, but it brings new management challenges: 1. How to manage multiple AI-driven development tasks simultaneously? 2. How to avoid context contamination between different tasks? 3. How to maintain supervision control when AI executes autonomously? 4. How to integrate AI workflows into existing Git and CI/CD processes? Agman (Agent Manager), built on Rust and ratatui, aims to address these issues. ## Core Design & Features: 1:1:1 Isolation Model and Multi-Dimensional Support ### 1:1:1 Task Isolation Model Each AI task corresponds to one Git branch, one Git worktree, and one tmux session, achieving full isolation. The advantages include no context switching cost, support for parallel development, clear task boundaries, and a safe experimental environment. ### Core Features - **Task Management**: Interactive creation wizard, status tracking, feedback mechanism, restart recovery; - **Agent Orchestration**: YAML-defined workflows, supporting reusable templates, auditability, extensibility, and conditional control; - **Git Worktree Integration**: Automatic worktree/branch management, support for branch strategies, change tracking; - **Tmux Integration**: Exclusive session for each task, including windows for nvim, lazygit, claude, etc.; - **GitHub Integration**: Automated PR operations (draft PR, CI monitoring, review tracking, etc.) and pre-packaged workflows. ## Technical Implementation: Building an Efficient TUI with Rust + ratatui Agman chooses Rust as its implementation language, paired with the ratatui framework to build the TUI: - **Rust Advantages**: Zero-cost abstractions, memory safety ensuring performance and reliability, rich CLI ecosystem accelerating development; - **ratatui Framework**: Declarative UI components, efficient terminal rendering, keyboard-driven interaction, cross-platform compatibility; - **Interaction Design**: Drawing on Vim philosophy, fully keyboard-operated, with preview panels, built-in editor, and shortcuts forming an efficient workflow. ## Security Considerations: Isolation Boundaries with 'Reckless by Design' Agman has a security warning: "Agman is reckless by design". All Claude Agents run in the `--dangerously-skip-permissions` mode, allowing AI to directly read/write files and execute commands. This design aims to maximize AI autonomy in a controlled isolated environment, with Git worktree isolation providing a security boundary (errors only affect the current task directory). However, it is not suitable for production environments, code repositories containing sensitive credentials, or scenarios where data loss cannot be tolerated. Developers should understand the security model before using it. ## Application Scenarios and Practical Value Agman is suitable for various AI-assisted development scenarios: - **Parallel Feature Development**: Multi-task parallelism shortens cycles; - **Code Refactoring**: Decomposing large tasks reduces risks; - **Test-Driven Development**: Automatically executing the "test-implement-fix" cycle; - **Code Review Assistance**: AI helps analyze PR issues; - **Exploratory Programming**: Trying new solutions in an isolated environment, discard if not satisfied. ## Limitations and Future Outlook ### Current Limitations - **Platform Restrictions**: Only supports macOS; Linux is theoretically usable but untested; Windows is not supported; - **Security Model**: Reckless mode has risks and is not suitable for all scenarios; - **Ecosystem Dependencies**: Relies on tools like tmux and nvim, with a high configuration threshold. ### Future Direction Agman represents the evolution of AI-assisted development tools from code completion to full workflow orchestration. As AI capabilities enhance, such tools will become more important in software development. ## Summary and Recommendations Agman is an innovation in development tools for the AI era. Its core design philosophy is to maximize AI autonomy in an isolated environment, achieving task isolation via Git worktrees and tmux, orchestrating workflows with YAML, and providing an efficient experience through TUI. Despite platform restrictions and security constraints, it provides a reference for future AI-assisted tools. It is recommended that developers already using AI assistants like Claude Code try Agman; it may change the way you collaborate with AI, turning multiple AI agents into a virtual development team.