agentsclimarketplace

Agent swe team

Skill northseadl/norix-skills/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).From its SKILL.md

Install
npx -y skills add northseadl/norix-skills --skill agent-swe-team

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 6 stars6 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

5.1 KB, ~1.6k tokens by cl100k_base, as published. Nobody here has run it

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

# 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:

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

$WS signal   # → "COMPLETED"
cd <项目目录>
RUN_ID=$($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 idleWorkers 完成但无动作$WS wake leader

CLI 工具

ws.mjs — 紧凑子命令

$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

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
--cwdcwd项目目录
--rolesleader,worker:2,inspector团队组成
--enginecodexcodex(thread resume) / claude(new session)
--baseHEAD基准 commit
--port自动端口号
--dry-runfalse模拟(不启动 Agent)

参考文档

文档用途何时读取
api_reference.mdHub 全部 HTTP API + JSON Schema需要直接调 HTTP API 时
internal_roles.mdLeader/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)

What ships with it: 19 files

143.4 KB alongside SKILL.md, 14 of them executable

references/

scripts/

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.