Team Collaboration Skill

快速搭建多 Agent 协作系统。创建产品/研发/运营团队，支持持久化、任务路由、知识提取、并行协作。

Security Scan

This is an instruction-only multi-agent collaboration template that is internally consistent: it reads/writes local memory files and spawns child agents but does not request credentials, installs, or external downloads by itself.

Purpose & Capability

The name/description (multi-agent team collaboration) matches the contents: templates and instructions to persist state, route tasks, extract knowledge, and spawn agents. No unrelated credentials, binaries, or install steps are required.

Instruction Scope

SKILL.md explicitly instructs reading and updating files under memory/ (MEMORY.md, company.md, doctor-profile.md) and spawning agents with state. This is coherent with the skill purpose. Note: the instructions encourage persisting user preferences/decisions and automatically greeting users (HEARTBEAT) if inactive — that implies the agent will store and act on user data and may send messages autonomously when invoked.

Install Mechanism

No install spec and no code files — instruction-only. This minimizes installation risk (nothing downloaded or written by an installer).

Credentials

The skill declares no required env vars or credentials (proportionate). However templates reference third‑party tools (GitHub CLI, notion, feishu, xhs, search tools) in examples; using those would require adding credentials later. The skill itself does not request or exfiltrate secrets.

Persistence & Privilege

The skill expects to read and write persistent files under memory/ (agent state, lessons, decisions). always:false and normal autonomous invocation flags are used. Persisting user preferences and decisions is expected for this feature, but users should be aware these files may contain sensitive info and the HEARTBEAT guidance schedules periodic checks/actions (every 30 minutes/24h greeting) which could trigger unsolicited messages when enabled.

Assessment

This skill appears to do what it says: templates + runtime instructions to spawn agents and store state locally. Before installing, confirm: (1) where the memory/ directory will be stored and who can read it (it will contain user preferences, decisions, and possibly sensitive snippets); (2) whether your agent platform implements the read()/spawnAgent() primitives used in the examples; (3) you are comfortable with the HEARTBEAT behavior that can send greetings/checks automatically; (4) you will only enable any third-party integrations (GitHub, Notion, Feishu, xhs, etc.) intentionally and supply credentials yourself. If you need stronger privacy, ask how to inspect, export, or delete the memory/ files and disable automatic greetings or scheduled checks.

快速搭建多 Agent 协作系统，让你的 AI 助手变成一个有组织的团队。

一键安装

skillhub install team-collab-skill

功能特性

功能	说明
Agent 持久化	状态文件系统，agent 记住上次在干什么
共享知识库	所有 agent 可读取的 Doctor 偏好、决策记录
任务自动路由	根据关键词自动分发给对应 agent
知识主动提取	自动识别关键信息并记录
监控告警	HEARTBEAT 定期检查
并行协作	多 agent 同时工作

文件结构

memory/
├── MEMORY.md              # 长期记忆
├── company.md             # 公司架构
├── lessons.md             # 学习记录
├── team-dashboard.md      # 团队仪表盘
├── shared/
│   ├── doctor-profile.md  # Doctor 偏好
│   ├── decisions.md       # 历史决策
│   ├── best-practices.md  # 最佳实践
│   ├── workflow.md        # 协作流程
│   ├── task-routing.md    # 任务路由
│   ├── knowledge-extraction.md  # 知识提取
│   └── parallel-collab.md # 并行协作
└── agents/
    ├── product-agent.md   # 产品团队
    ├── dev-agent.md       # 研发团队
    └── ops-agent.md       # 运营团队

使用方法

1. 初始化

在 Agent 启动时，读取以下文件：

• memory/MEMORY.md — 长期记忆
• memory/company.md — 公司架构
• memory/shared/doctor-profile.md — Doctor 偏好

2. Spawn Agent

创建子 agent 时，传入状态文件：

// 读取状态
const state = read('memory/agents/product-agent.md');

// Spawn
spawnAgent('product', `
你是产品团队 agent。

## 你的状态
${state}

## 任务
${task}
`);

3. 任务路由

根据关键词分发任务：

关键词	分发给
规划、分析、需求、功能	产品 Agent
安装、技能、代码、配置	研发 Agent
小红书、文案、图片、数据	运营 Agent

4. 知识提取

自动识别 Doctor 说的关键信息：

Doctor 说	提取	存到
"我是..."	身份	doctor-profile.md
"我喜欢..."	偏好	doctor-profile.md
"记住..."	决策	decisions.md
"不要再犯..."	教训	lessons.md

5. 并行协作

多个 agent 同时工作：

spawnAgent('product', '任务A');
spawnAgent('dev', '任务B');
spawnAgent('ops', '任务C');

// 等待所有完成
await Promise.all([...]);

模板文件

所有模板在 templates/ 目录：

• agents/*.md — Agent 状态模板
• shared/*.md — 共享知识模板

最佳实践

1. 每次 spawn 前读取状态文件
2. 任务完成后更新状态文件
3. 重要发现写入 lessons.md
4. 定期检查 HEARTBEAT.md

示例

Doctor: "帮我规划一个功能"

微³:

1. 读取 product-agent.md
2. Spawn 产品 agent
3. 产品输出规划
4. 更新 product-agent.md
5. 汇报给 Doctor

---

Created by 微³ 🦞 龙虾人科技