什么是 Agent Squad？

Agent Squad 是一个强大的编排层，旨在管理在后台运行的持久 AI 编码团队。通过利用 tmux 会话，此技能允许 AI 智能体在复杂的编码任务上全天候 24/7 工作，而无需用户保持终端窗口打开。它是 Openclaw Skills 用户将长期运行的开发工作委托给自主智能体的核心组件。

该技能支持多种 AI 引擎，包括 Claude Code、Aider 和 Gemini，提供统一的界面来同时启动、停止和监控多个小组。它具有自动化看门狗系统，可确保您的智能体保持健康和高效，在遇到错误或崩溃时自动重启它们。

下载入口:https://github.com/openclaw/skills/tree/main/skills/0xtimi/agent-squad

安装与下载

1. ClawHub CLI

从源直接安装技能的最快方式。

2. 手动安装

将技能文件夹复制到以下位置之一

优先级：工作区 > 本地 > 内置

3. 提示词安装

将此提示词复制到 OpenClaw 即可自动安装。

Agent Squad 应用场景

• 在您专注于其他任务时，在后台运行大规模代码库重构。
• 编排多个 AI 引擎在同一个项目中的不同微服务上工作。
• 维护持续开发循环，AI 智能体在任务分配后立即从队列中提取任务。
• 通过实时仪表板和实时终端查看功能监控自主智能体的进度。

1. 用户创建一个命名小组并分配特定的 AI 引擎和项目目录。
2. Agent Squad 初始化一个专用的 tmux 会话来托管智能体，确保进程即使在用户断开连接后也能持久运行。
3. 通过斜杠命令或自然语言输入将任务添加到持久队列中。
4. 智能体按顺序处理任务，在结构化工作区中生成 Markdown 报告和日志。
5. 后台看门狗进程每 5 分钟检查每个小组的状态，如果发现会话处于非活动状态，则执行自动重启。
6. 用户通过仪表板与小组交互，该仪表板从智能体的实时终端输出和状态报告中提取实时数据。

Agent Squad 配置指南

要开始使用此 Openclaw Skills 技能，请确保您的系统中安装了 tmux 和 python3。您可以使用以下命令启动您的第一个编码团队：

/agent-squad start my-first-squad claude --project ~/path/to/your/code

小组运行后，您可以立即开始分配开发任务：

/agent-squad assign my-first-squad "Create a React component for the user dashboard"

Agent Squad 数据架构与分类体系

Agent Squad 在 Openclaw 环境中为每个团队维护一个结构化目录，以确保数据持久性和易于审计。

name: agent-squad
version: 0.9.4
license: MIT-0
description: "Manage persistent AI coding squads that run in tmux sessions with task queues, progress reports, and automatic health monitoring. Use when the user wants to: (1) start/launch/create/restart a squad or team of AI agents, (2) assign/give tasks to a squad, (3) check squad status or ask what a squad is doing, (4) ping/nudge a squad to report progress, (5) stop a squad, (6) list all active squads, (7) configure squad settings like default project directory, (8) delete/archive a squad. Supports Claude Code, Codex, Gemini CLI, OpenCode, Kimi, Trae, Aider, and Goose as AI engines."
metadata:
  { "openclaw": { "requires": { "anyBins": ["tmux"], "bins": ["python3"] } } }

Agent Squad

GitHub: https://github.com/0xTimi/agent-squad

Run persistent AI coding squads in tmux. Squads pick up tasks, write code, and report progress — 24/7 in the background.

Slash Command Usage

Users can invoke /agent-squad directly with optional arguments:

No arguments or list → run bash {baseDir}/scripts/squad-list.sh:

• If squads exist: show a clean status dashboard
• If no squads: show the Getting Started intro below

Getting Started

When users ask "what is this", "how do I use this", or invoke /agent-squad with no squads, give a friendly intro with usage examples. Match the user's language.

Agent Squad runs AI coding agents in the background 24/7. Just tell me what you need:

• "Start a squad called my-squad using claude for ~/projects/my-app"
• "Give my-squad a task: implement user login"
• "How's my-squad doing?"
• "Stop my-squad"
• "What squads do I have?"

Engines: Claude Code, Codex, Gemini CLI, OpenCode, Kimi, Trae, Aider, Goose

Want to start one now?

If the user asks which engine to use or doesn't specify one: default to claude — it is the most thoroughly tested engine and works best with Agent Squad. Mention other engines only if the user specifically asks. See {baseDir}/references/engines.md for details.

What Users Can Do

Users interact through natural language. Here's what they might say and how to respond:

Start a squad

User: "start a squad called my-squad with claude" / "launch a codex squad for ~/projects/api"

Ask if missing: squad name. Default engine to claude if not specified. Project dir and context are optional.

First-time users: briefly mention squads run in full-auto mode — the AI has full access to the project directory.

Response: "Squad 'my-squad' is up and running with Claude Code! You can assign tasks anytime."

Assign a task

User: "give my-squad a task: build the login page" / "let my-squad work on JWT auth"

If only one squad exists, use it automatically. If the request is vague, ask for specifics.

Response: "Task assigned! my-squad will start working on 'Login Page' shortly."

Check status

User: "how's my-squad doing?" / "what's the status?" / "is my-squad done yet?"

Include the live tmux screen output (from squad-status.sh) in the response — this lets the user see what the agent is actually doing right now.

Response: "my-squad is running on Claude Code, working on 'Login Page' — about 60% done. 2 tasks completed, 1 in progress.

Live screen:

Working on form validation...
Created src/components/LoginForm.tsx
Running tests...
```"

### Peek at screen

User: "peek at my-squad" / "what's on my-squad's screen?" / "show me what my-squad is doing"

Show the raw tmux screen content. This is a quick way to see the agent's live terminal without checking reports.

Response: Show the screen output directly, formatted in a code block.

### Ping for update

User: "ping my-squad" / "nudge it to report"

Response: "I've nudged my-squad to update its progress report. Check back in a minute."

### Stop a squad

User: "stop my-squad" / "pause the squad"

Always confirm before stopping.

Response: "my-squad stopped. All work is saved — you can restart anytime."

### Restart a squad

User: "restart my-squad" / "bring my-squad back up"

Response: "my-squad is back up and running! It will pick up where it left off."

### List squads

User: "what squads do I have?" / "list my squads" / "show all squads"

Present a clean readable summary of all squads with name, engine, status, and task counts.

### Delete a squad

User: "delete my-squad" / "archive the old squad" / "clean up my-squad"

Always ask for confirmation first. Reassure: data is archived, project code is never touched.

### Configure

User: "set default project dir to ~/code" / "show squad settings"

### What commands are available?

User: "agent-squad都有哪些命令" / "what can you do?" / "help"

**IMPORTANT**: Never show internal script names (squad-start.sh, etc.) to users. Instead, show the `/agent-squad` slash commands and natural language examples:

> Here's what you can do with Agent Squad:
>
> | Command | Or just say... |
> |---|---|
> | `/agent-squad start my-squad claude` | "Start a squad called my-squad" |
> | `/agent-squad assign my-squad "task"` | "Give my-squad a task: ..." |
> | `/agent-squad status my-squad` | "How's my-squad doing?" |
> | `/agent-squad peek my-squad` | "Peek at my-squad's screen" |
> | `/agent-squad ping my-squad` | "Ping my-squad" |
> | `/agent-squad stop my-squad` | "Stop my-squad" |
> | `/agent-squad restart my-squad` | "Restart my-squad" |
> | `/agent-squad list` | "What squads do I have?" |
> | `/agent-squad delete my-squad` | "Delete my-squad" |

---

## Script Reference

All scripts at `{baseDir}/scripts/`. Execute based on user intent above and present results conversationally.

### squad-start.sh

```bash
bash {baseDir}/scripts/squad-start.sh "" "" "" [--project ] [--restart] [--agent-teams] [--no-watchdog]

• name: lowercase alphanumeric + hyphens
• engine: claude, codex, gemini, opencode, kimi, trae, aider, goose
• context: optional project background
• --project : custom code output directory
• --restart: required if squad name already exists (also used for restart intent)
• --agent-teams: claude only, multi-agent mode
• --no-watchdog: skip auto-restart cron

squad-assign.sh

bash {baseDir}/scripts/squad-assign.sh "" "