Agent Team

v1.5.0智能体编排

opencode-agent-team

Agent Team plugin for OpenCode — role injection and quality checks

维护活跃

#opencode #opencode-plugin #agent-team #multi-agent #ai-team

GitHub 仓库 npm 页面官方主页

GitHub 星标

月装机量

近 7 天 8

综合评分SCORE

35.9

生态多维模型

快速安装与配置

opencode.json

写入当前项目的 opencode.json，只对这个仓库生效。

opencode.json

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-agent-team@1.5.0"]
}

opencode 启动时会通过内嵌运行时自动加载 npm 依赖并缓存至本地目录，无需手动在全局环境执行安装。

README.md

查看原始文件→

English | 中文

Orchestrate your AI workforce with surgical precision. 🚀

agent-team is a multi-agent development manager that uses a Role + Worker model to run AI agents in isolated Git worktrees and dedicated terminal sessions.

🎭 Role: Reusable skill packages (.agent-team/teams/) defining goals, prompts, and tools.
🛠️ Worker: Isolated runtime instances (.worktrees/) with their own branch and session.

Installation
- AI-Native (Recommended)
- Manual Installation
Upgrade
Quick Start
The Toolkit
Supported Providers
Advanced Usage
License

🛠️ Installation

🤖 AI-Native (Recommended)

Let your AI agent set itself up. It only takes two steps:

Install the Skill:
```
npx skills add JsonLee12138/agent-team -a <platform> -y
```
Replace <platform> with your provider: claude, gemini, opencode, or codex.
Tell your Agent:

"Install agent-team and initialize the project."

📦 Manual Installation

Method	Command
Homebrew	`brew tap JsonLee12138/agent-team && brew install agent-team`
Go Install	`go install github.com/JsonLee12138/agent-team@latest`
Claude Plugin	`/plugin marketplace add JsonLee12138/agent-team`
Gemini Ext	`gemini extensions install https://github.com/JsonLee12138/agent-team`
OpenCode Plugin	`{ "plugin": ["opencode-agent-team"] }` in `~/.config/opencode/opencode.json`

🔄 Upgrade

Method	Command
Claude Plugin	`/plugin marketplace update agent-team`
Skill	`npx skills add JsonLee12138/agent-team -a '*' -y`
Homebrew	`brew update && brew upgrade agent-team`
Go Install	`go install github.com/JsonLee12138/agent-team@latest`

🚀 Quick Start

Manage your entire team through natural language. Your AI agent will handle the commands for you.

1. Define your Team

"Create a frontend-architect role to manage our UI infrastructure."

2. Install Curated Roles (Optional)

Pull in expert roles from this repository:

agent-team role-repo add JsonLee12138/agent-team

3. Spawn a Worker

"Create a worker for frontend-architect using Claude." This opens a new terminal window with an isolated worktree.

4. Assign & Brainstorm

"Assign frontend-architect-001 to design the new auth flow." The agent will brainstorm a design doc before touching any code.

5. Merge & Cleanup

"Merge frontend-architect-001 and delete the worker."

🧰 The Toolkit

Built-in Roles

Available in .agent-team/teams/:

pm: Product Manager & Requirement Shaping.
frontend-architect: High-level UI/UX structure.
vite-react-dev: Specialized for Vite + React.
pencil-designer: UI design tool specialist.

Built-in Skills

task-orchestrator: Task lifecycle entry.
workflow-orchestrator: Governance-only workflow plan entry.
worker-dispatch: Controller-side worker open/reply entry.
worker-recovery: Worker-first resume entry.
worker-reply-main: Worker -> main reporting entry.
context-cleanup: Session cleanup + index-first file re-anchoring.
task-inspector: Read-only task inspection.
role-repo-manager: Role source management.
catalog-browser: Read-only catalog browsing.
project-bootstrap: init / migrate entry.
rules-maintenance: rules sync entry.
skill-maintenance: Skill cache maintenance.
worker-inspector: Read-only worker status.
role-browser: Read-only local role browsing.
role-creator: Interactively build new agent roles.
brainstorming: Validates ideas via dialogue before implementation, including planning-target selection (roadmap / milestone / phase / task / generic topic) and explicit save-destination choices (docs/brainstorming/, target-object directory, custom path, or skip saving).
agent-team: Legacy compatibility shell that now routes to the dedicated scenario skills.

🤖 Supported Providers

Provider	CLI Value	Integration
Claude Code	`claude`	Plugin
Gemini CLI	`gemini`	Extension
OpenCode	`opencode`	NPM Plugin
OpenAI Codex	`codex`	Prompt-only

⚙️ Advanced Usage

📖 CLI Reference

Role Management

agent-team role list: Show local roles.
agent-team role create <name>: Create a new role package (SKILL.md, references/role.yaml, system.md) under skills/, .agent-team/teams/, or a custom target path.
agent-team role-repo add <owner/repo>: Install roles from GitHub.

Worker Operations

agent-team worker create <role> [--provider <provider>] [--model <model>]: Prepare a new worker (does not start a session).
agent-team worker open <worker-id> [--provider <provider>] [--model <model>] [--new-window]: Start or reopen a worker session.
agent-team worker close <worker-id>: Close a worker session without deleting the worker.
agent-team worker status: View active workers and tasks.
agent-team worker assign <id> "<task>": Dispatch work.
agent-team worker merge <id>: Sync worker changes back (does not close the session).
agent-team worker delete <id>: Remove a worker and its worktree.

Communication

agent-team reply <id> "<msg>": Send message to worker.
agent-team reply-main "<msg>": Worker talks back to main.

Planning Artifacts

agent-team planning create --kind <roadmap|milestone|phase> "<title>": Create a planning artifact.
agent-team planning list [--kind <kind>] [--lifecycle <planning|archived|deprecated>]: List planning artifacts.
agent-team planning show <id>: Show a planning artifact and reference checks.
agent-team planning move <id> --to <planning|archived|deprecated>: Move a planning artifact across lifecycle directories.

Task Artifacts

Every active task package contains three standard artifacts under .agent-team/task/<task-id>/:

task.yaml: lifecycle metadata and status.
context.md: background, scope, constraints, and design input.
verification.md: acceptance contract, test scope boundary, performed checks, and final verification result.

verification.md is created automatically with agent-team task create. Its default template keeps E2E Required: no, Verified By: qa, and a pending result so QA or human acceptance can complete the record later.

Lifecycle summary:

task done now moves a task from assigned to verifying.
task archive reads verification.md and only archives tasks whose ## Result passes the gate.
task deprecated moves unfinished or abandoned work into .agent-team/deprecated/task/<task-id>/ while preserving the task package.

📂 Directory Structure

project-root/
├── .agent-team/task/                  <- Active task packages
├── .agent-team/planning/roadmaps/     <- Active roadmap artifacts
├── .agent-team/planning/milestones/   <- Active milestone artifacts
├── .agent-team/planning/phases/       <- Active phase artifacts
├── .agent-team/archive/task/          <- Archived task packages
├── .agent-team/archive/roadmaps/      <- Archived roadmap artifacts
├── .agent-team/archive/milestones/    <- Archived milestone artifacts
├── .agent-team/archive/phases/        <- Archived phase artifacts
├── .agent-team/deprecated/task/       <- Deprecated task packages
├── .agent-team/deprecated/roadmaps/   <- Deprecated roadmap artifacts
├── .agent-team/deprecated/milestones/ <- Deprecated milestone artifacts
├── .agent-team/deprecated/phases/     <- Deprecated phase artifacts
├── .agent-team/teams/                 <- Project-specific roles
├── .worktrees/                        <- Isolated worker workspaces
├── roles-lock.json                    <- Remote role version locking
└── gemini-extension.json              <- Extension manifest

🌐 Environment Variables

Variable	Default	Description
`AGENT_TEAM_BACKEND`	`wezterm`	Terminal: `wezterm` or `tmux`.
`AGENT_TEAM_ROLE_HUB_URL`	`https://...`	Ingest endpoint for analytics.
`AGENT_TEAM_ROLE_HUB_DEBUG`	`0`	Wait for ingest if set to `1`.