agent-swe-team
Multi-agent SWE team built on the Workshop model. Full-stack vertical workers, meeting room with @mention notification, private pipes, shared task board. Git worktree isolation, Leader-driven coordination. Mixed Codex/Claude Code engine. Use when a task needs engineering depth beyond a single agent. NOT for simple task parallelism (use agent-task-orchestration) or design discussions (use agent-brainstorm).
Best use case
agent-swe-team is best used when you need a repeatable AI agent workflow instead of a one-off prompt.
Multi-agent SWE team built on the Workshop model. Full-stack vertical workers, meeting room with @mention notification, private pipes, shared task board. Git worktree isolation, Leader-driven coordination. Mixed Codex/Claude Code engine. Use when a task needs engineering depth beyond a single agent. NOT for simple task parallelism (use agent-task-orchestration) or design discussions (use agent-brainstorm).
Teams using agent-swe-team should expect a more consistent output, faster repeated execution, less prompt rewriting.
When to use this skill
- You want a reusable workflow that can be run more than once with consistent structure.
When not to use this skill
- You only need a quick one-off answer and do not need a reusable workflow.
- You cannot install or maintain the underlying files, dependencies, or repository context.
Installation
Claude Code / Cursor / Codex
Manual Installation
- Download SKILL.md from GitHub
- Place it in
.claude/skills/agent-swe-team/SKILL.mdinside your project - Restart your AI agent — it will auto-discover the skill
How agent-swe-team Compares
| Feature / Agent | agent-swe-team | Standard Approach |
|---|---|---|
| Platform Support | Not specified | Limited / Varies |
| Context Awareness | High | Baseline |
| Installation Complexity | Unknown | N/A |
Frequently Asked Questions
What does this skill do?
Multi-agent SWE team built on the Workshop model. Full-stack vertical workers, meeting room with @mention notification, private pipes, shared task board. Git worktree isolation, Leader-driven coordination. Mixed Codex/Claude Code engine. Use when a task needs engineering depth beyond a single agent. NOT for simple task parallelism (use agent-task-orchestration) or design discussions (use agent-brainstorm).
Where can I find the source code?
You can find the source code on GitHub using the link provided at the top of the page.
Related Guides
SKILL.md Source
# Agent SWE Team — Workshop Engine
## 核心身份 — Supervisor (监工)
你是**用户的持续在线代理**。当用户给出工程目标并触发本技能,你**成为 Workshop Supervisor**。
你不写代码——你启动一支全栈工程团队,监控进度,转发人类意图,保障系统透明。
**三层架构**:
```
用户 ─── 给出目标 ─── 可随时介入
│
▼
你 (Supervisor) ── 启动 Workshop → 监控 → 转发 → 收尾
│
▼
Hub (HTTP Server) ── 纯管道 + @mention 自动唤醒
│
├── Leader ── 内部协调者:分解目标、分配任务、质检、收工
├── Worker×N ── 全栈工匠:各自独立 worktree,垂直切片
└── Inspector ── 质检官:基于原始目标整体评估
```
**行为准则**:
- **你是最后防线**。Hub 进程在后台运行,没有你监控就不透明。
- **你不做内部决策**。任务分解、分配、质检由 Leader 处理。
- **你持续轮询**。以退避节奏监控状态。
- **你转发人类意图**。用户说什么 → `ws say "消息"`。
## 全流程
### Phase 1: 启动
```bash
# 1. 确保依赖(只需首次)
cd <skill-dir>/agent-swe-team && npm install
# 2. 后台启动(serve 不返回)
node <skill-dir>/agent-swe-team/scripts/team.mjs serve \
--cwd <项目目录> --goal "你的目标" &
# 3. 等待就绪 + 获取端口
sleep 3
PORT=$(cat <项目目录>/.workshop/port)
export WORKSHOP_CWD=<项目目录>
```
Hub 启动后自动:创建 `.workshop/`、integration 分支、Worker worktrees、打开 Dashboard、唤醒 Leader。
### Phase 2: 监控循环
使用 `ws.mjs` CLI 工具减少上下文消耗:
```bash
WS="node <skill-dir>/agent-swe-team/scripts/ws.mjs"
```
**退避轮询**:
```
Phase 间隔 命令
启动确认 60s $WS signal
运行中 120s $WS signal → 如果 RUNNING → $WS board
尾声 300s $WS signal → 等 COMPLETED
```
```
$WS signal
├─ "COMPLETED" → 跳到 Phase 3
└─ "RUNNING" → $WS board
if leader idle 且有 Worker 完成/空闲:
$WS wake leader
```
**人类消息**: `$WS say "用户说的内容"`
**@mention 自动唤醒**: 内部 Agent 在会议室 @另一个 Agent 时,Hub 自动唤醒被提及者并注入新消息。Supervisor 无需干预。
### Phase 3: 收尾
```bash
$WS signal # → "COMPLETED"
cd <项目目录>
RUN_ID=$($WS board | head -1 | awk '{print $NF}')
git diff main..integration/$RUN_ID --stat
git checkout main && git merge integration/$RUN_ID --no-ff
```
### 错误恢复
| 场景 | 信号 | 处理 |
|:---|:---|:---|
| Worker 异常 | 会议室 `"异常终止"` | `$WS wake leader` |
| merge 冲突 | 会议室 `"合并失败"` | `$WS wake leader` |
| Hub 进程挂 | PORT 无响应 | 重新 `serve`(board.json 恢复) |
| Leader 卡 idle | Workers 完成但无动作 | `$WS wake leader` |
## CLI 工具
### ws.mjs — 紧凑子命令
```bash
$WS signal # → "RUNNING" 或 "COMPLETED"
$WS board # 紧凑面板视图 (~10 行)
$WS wake leader # 唤醒 Agent
$WS say "message" # 发到会议室
$WS say "@worker-1 检查一下日志" # @mention → 自动唤醒 worker-1
$WS dm worker-1 "私信" # 发 DM
$WS task create "标题" --assign worker-1 # 创建任务
$WS task complete 1 "摘要" # 完成任务
$WS task progress 1 50 "备注" # 更新进度
$WS merge worker-1 # 合并分支
$WS done # 结束运行
$WS chat # 读会议室
```
### team.mjs — 启动命令
```bash
node scripts/team.mjs serve --goal "目标" [选项]
node scripts/team.mjs status [--cwd <DIR>]
```
| 选项 | 默认值 | 说明 |
|:---|:---|:---|
| `--goal` | 必需 | 目标描述 |
| `--cwd` | cwd | 项目目录 |
| `--roles` | `leader,worker:2,inspector` | 团队组成 |
| `--engine` | `codex` | `codex`(thread resume) / `claude`(new session) |
| `--base` | `HEAD` | 基准 commit |
| `--port` | 自动 | 端口号 |
| `--dry-run` | false | 模拟(不启动 Agent) |
## 参考文档
| 文档 | 用途 | 何时读取 |
|:---|:---|:---|
| [api_reference.md](references/api_reference.md) | Hub 全部 HTTP API + JSON Schema | 需要直接调 HTTP API 时 |
| [internal_roles.md](references/internal_roles.md) | Leader/Worker/Inspector 内部机制 | 理解内部行为 / 调试时 |
## 约定
- **规划/汇报**: 中文
- **代码/命令/文件名**: English
- **Git commit**: 中文 (Conventional Commits)Related Skills
web-scraper
Web scraper with SPA/JavaScript rendering, page interaction, and JS execution. Two-tier engine (HTTP → Playwright browser). Smart discovery, batch fetch, interactive content extraction, OpenAPI parsing. Use when read_url_content fails, SPA rendering needed, or page interaction required.
pm-toolkit
自然语言 → Mermaid 图表(16 种)+ 网页原型,本地 Web 面板实时预览/导出。
mobile-testing
Android/iOS automated testing: device management, app install, performance profiling, log analysis, screenshot comparison, Maestro E2E orchestration.
llm-agent-dev
LLM Agent engineering: pattern selection (12-mode matrix), data simulation, convergence iteration. Covers intent routing, function calling, ReAct, MCP, prompt chaining, guardrails, evaluation.
image-studio
AI image generation and editing: e-commerce templates (hero/banner/detail/lifestyle), image refinement (background replace/remove, enhance, retouch, style transfer), icon extraction (bg-removal + detect + smart crop with transparent output).
feishu-integration
Feishu (Lark) unified CLI for tasks, documents, wiki, bitable, messaging, approval, and Drive. Supports search/create/edit/publish/export across all modules. Use when reading/writing Feishu docs, searching docs or wiki, managing tasks, sending messages, creating approvals, exporting to Markdown, or any 飞书/Lark interaction.
es-analytics
Elasticsearch / SLS 只读数据分析:索引探索、mapping、聚合统计、日志搜索、时间序列、多 Profile。
doc-sentinel
Document-code change notification system: traceable doc-code binding via git tree hash, git-diff-driven reconciliation plans with confidence/risk metadata, and idempotent execution. Use when maintaining documentation freshness, detecting stale docs, or binding docs to source code.
coding-net-integration
Coding.net DevOps automation: MR lifecycle, CI operations (trigger/logs/stop), artifact registry, cross-project queries, remote file audit.
cnb-cool-integration
针对 cnb.cool 的云原生构建(CNB Build):生成/修改 .cnb.yml 流水线、触发规则、 构建环境、runner 资源、缓存、环境变量、手动触发与调试。 在接入、迁移、优化或排查 CNB 构建配置时使用。
agent-task-orchestration
Task decomposition and multi-agent orchestration with retry, checkpoint recovery, and real-time monitoring. Mixed Codex/Claude Code engine. Parallel/sequential execution.
agent-front-design
Frontend design blueprints with craftsmanship scoring, self-critique loops, and engineering handoff. Aesthetic intelligence against AI homogeneity. Use for: UI/UX design specs, design system creation, visual direction exploration, component design review, design-to-engineering handoff.