Agency Agents

AI Agent 团队 - 61 个专业 Agent,8 大部门,完整的 AI 代理机构。支持单 Agent 使用和多 Agent 协作编排。

MIT-0 · Free to use, modify, and redistribute. No attribution required.
0 · 104 · 0 current installs · 0 all-time installs
MIT-0
Security Scan
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Benign
medium confidence
Purpose & Capability
The name/description (an 'agency' of specialized agents) matches the content: many persona documents and an orchestrator SKILL.md. There are no unexpected required binaries or credentials listed; the files are primarily documentation and examples, which is coherent with the stated purpose.
Instruction Scope
The runtime SKILL.md is documentation-only and instructs use via /openclaw skill use agency-agents commands. It also documents optional environment variables and a local output path (~/clawd/agency-agents/...). It does not instruct reading arbitrary system files or exfiltrating data, but some agent docs include examples (e.g., docker-compose with JWT_SECRET, DATABASE_URL) that could lead an agent to request secrets during real tasks — this is expected for agents that integrate with external systems, but it's out-of-band from the skill's declared requirements.
Install Mechanism
No install spec or code execution/install scripts are present; the package is instruction-only. README suggests optional git clone from a 'your-repo' placeholder, but there is no automated download-from-URL or archive extraction to flag.
Credentials
The registry lists no required environment variables or credentials. SKILL.md documents optional env vars (AGENCY_AGENTS_*) and some example config snippets reference secrets (JWT_SECRET, DATABASE_URL). Because none are required up front, the request surface is proportional; nonetheless, be careful to avoid providing any sensitive credentials unless you explicitly intend an Agent to access an external service.
Persistence & Privilege
The skill is not always:true and uses the default autonomous-invocation setting (disable-model-invocation: false). There is no install-time persistence or modifications to other skills. This level of privilege is normal for an instruction-only skill; combine with other concerns only if you plan to let it run autonomously with sensitive data.
Scan Findings in Context
[no_findings] expected: The regex-based scanner had nothing to analyze because this is an instruction-only skill with no executable code files. That's expected for a docs/persona package; absence of findings is not proof of safety — review the instructions and examples before use.
Assessment
Plain-language considerations before installing/using: - This skill is documentation-only (61 persona files + orchestrator docs). It does not install binaries or require credentials up front, which is consistent with its purpose. - There are minor metadata inconsistencies (different ownerId/version in _meta.json and registry) and placeholder links (e.g., git URLs like 'your-repo') — these suggest the package may be a locally adapted fork. Verify the publisher/source and prefer published repositories with an identifiable maintainer before trusting it in production. - The docs include examples that reference environment variables and secrets (DATABASE_URL, JWT_SECRET). Do not enter or expose production credentials to any Agent unless you explicitly intend to connect that Agent to a service and trust the provenance. - The skill will save outputs to a local path (docs mention ~/clawd/agency-agents/) — if you care about where outputs land, review and, if needed, run in a sandbox or with an isolated workspace. - If you plan to allow autonomous or orchestrated runs (the default), test interactions with non-sensitive sample data first and review orchestrator/SKILL.md to understand what orchestration steps the orchestrator persona will take. If you want higher confidence, ask the publisher for: 1) a canonical source repository or homepage and an explanation of the ownerId/version discrepancies; 2) the orchestrator/SKILL.md runtime behavior (it wasn't fully shown in the registry payload) so you can verify no hidden network callbacks; and 3) explicit notes about what agents may request (credentials, external API tokens) during typical tasks.

Like a lobster shell, security has layers — review code before you run it.

Current versionv1.1.0
Download zip
latestvk975xqrbjt8g6pzgmd3q2hk431834h6d

License

MIT-0
Free to use, modify, and redistribute. No attribution required.

SKILL.md

AI Agent 团队 - OpenClaw 技能

📦 技能概述

本技能提供 61 个专业 AI Agent,分为 8 大业务部门,可独立完成专业任务或多 Agent 协作完成复杂项目。

🎯 使用方式

方式 1:使用单个 Agent

# 语法
/openclaw skill use agency-agents --agent <agent-name> "<任务描述>"

# 示例
/openclaw skill use agency-agents --agent frontend-developer "帮我创建一个 React 登录页面组件"
/openclaw skill use agency-agents --agent growth-hacker "为我的 SaaS 产品设计增长策略"
/openclaw skill use agency-agents --agent ui-designer "审查我的网站设计并提供改进建议"

方式 2:使用 Agent 编排器(推荐)

# 自动调度多个 Agent 完成复杂项目
/openclaw skill use agency-agents --agent orchestrator "开发一个完整的电商网站,包括前端、后端和营销策略"

方式 3:使用整个部门

# 激活整个部门的 Agent 协作
/openclaw skill use agency-agents --department engineering "开发一个完整的 Web 应用"
/openclaw skill use agency-agents --department marketing "策划并执行一次营销活动"

🏢 可用部门

部门Agent 数量用途
engineering7工程开发
design7设计相关
marketing8市场营销
product3产品管理
project-management5项目管理
testing7测试 QA
support6业务支持
specialized6专业领域

🎭 可用 Agent

💻 工程部

  • frontend-developer - 前端开发专家(React/Vue/Angular)
  • backend-architect - 后端架构师(API/数据库/云)
  • mobile-app-builder - 移动端开发(iOS/Android/跨平台)
  • ai-engineer - AI 工程师(ML/深度学习/AI 集成)
  • devops-automator - DevOps 自动化(CI/CD/基础设施)
  • rapid-prototyper - 快速原型(MVP/POC)
  • senior-developer - 高级开发(复杂实现/架构决策)

🎨 设计部

  • ui-designer - UI 设计师(视觉/组件/设计系统)
  • ux-researcher - UX 研究员(用户研究/可用性测试)
  • ux-architect - UX 架构师(技术架构/CSS 系统)
  • brand-guardian - 品牌守护者(品牌策略/一致性)
  • visual-storyteller - 视觉叙事者(视觉故事/多媒体)
  • whimsy-injector - 趣味注入师(微交互/品牌个性)
  • image-prompt-engineer - 图像提示工程师(AI 图像生成)

📢 市场部

  • growth-hacker - 增长黑客(用户获取/转化优化)
  • content-creator - 内容创作者(多平台内容/编辑日历)
  • twitter-engager - Twitter 互动专家(实时互动/思想领导)
  • tiktok-strategist - TikTok 策略师(病毒内容/算法优化)
  • instagram-curator - Instagram 策划师(视觉叙事/社群)
  • reddit-community-builder - Reddit 社群建设者(社群运营)
  • app-store-optimizer - 应用商店优化师(ASO/转化)
  • social-media-strategist - 社交媒体策略师(跨平台策略)

📊 产品部

  • sprint-prioritizer - 冲刺优先级专家(敏捷规划)
  • trend-researcher - 趋势研究员(市场情报/竞争分析)
  • feedback-synthesizer - 反馈整合师(用户反馈分析)

🎬 项目管理部

  • studio-producer - 工作室制作人(高层协调/组合管理)
  • project-shepherd - 项目协调员(跨职能协调)
  • studio-operations - 工作室运营(日常效率/流程优化)
  • experiment-tracker - 实验追踪师(A/B 测试)
  • senior-pm - 高级项目经理(范围规划/任务分解)

🧪 测试部

  • evidence-collector - 证据收集师(截图 QA/视觉验证)
  • reality-checker - 现实检查员(质量认证/发布审批)
  • test-results-analyzer - 测试结果分析师(测试评估)
  • performance-benchmarker - 性能基准师(性能测试)
  • api-tester - API 测试师(API 验证/集成测试)
  • tool-evaluator - 工具评估师(技术评估)
  • workflow-optimizer - 工作流优化师(流程分析)

🛟 支持部

  • support-responder - 客服响应师(客户服务)
  • analytics-reporter - 分析报告师(数据分析/仪表板)
  • finance-tracker - 财务追踪师(财务规划/预算)
  • infrastructure-maintainer - 基础设施维护师(系统可靠性)
  • legal-compliance-checker - 法律合规师(合规/法规)
  • executive-summary-generator - 高管摘要生成师(C -suite 沟通)

🎯 特别部门

  • orchestrator - Agent 编排器(多 Agent 调度核心)
  • data-analytics-reporter - 数据分析报告师(商业智能)
  • lsp-index-engineer - LSP/索引工程师(代码智能)
  • sales-data-extraction-agent - 销售数据提取师
  • data-consolidation-agent - 数据整合师
  • report-distribution-agent - 报告分发师

🔄 核心工作流

单 Agent 工作流

1. 用户指定 Agent 和任务
2. 加载对应 Agent 人格和专业知识
3. 执行任务并输出专业结果
4. 提供可交付成果

多 Agent 编排工作流

1. 项目分析阶段
   └─→ 高级项目经理:任务分解
   
2. 架构设计阶段
   └─→ 架构师:技术架构设计
   
3. 开发实施阶段(循环)
   ├─→ 开发工程师:实现任务
   └─→ 测试工程师:质量验证
   └─→ (如失败)→ 返回开发重试
   
4. 集成测试阶段
   └─→ 现实检查员:最终质量认证
   
5. 交付阶段
   └─→ 生成完整交付文档

📋 输出格式

标准输出

# [Agent 名称] - [任务类型]

## 🎯 任务理解
[对任务的理解和分析]

## 📋 执行过程
[详细的工作步骤]

## 📦 交付成果
[具体的交付物,如代码、文档、策略等]

## ✅ 质量检查
[自我验证和质量保证]

## 💡 建议
[后续优化建议]

编排器输出

# 项目完成报告

## 📊 项目概览
- 总任务数:X
- 完成时间:Y
- 参与 Agent:Z

## ✅ 完成情况
[所有任务的完成状态]

## 📦 交付清单
[所有交付物的列表和链接]

## 📈 质量报告
[整体质量评估和各项指标]

⚙️ 配置选项

环境变量

# 设置默认部门
AGENCY_AGENTS_DEFAULT_DEPARTMENT=engineering

# 设置质量检查严格度(1-5)
AGENCY_AGENTS_QA_LEVEL=3

# 设置最大重试次数
AGENCY_AGENTS_MAX_RETRIES=3

# 启用详细日志
AGENCY_AGENTS_VERBOSE=true

🎯 最佳实践

1. 选择合适的 Agent

  • 明确任务类型 → 选择对应专业 Agent
  • 复杂项目 → 使用编排器
  • 需要多领域 → 使用部门协作

2. 提供清晰的任务描述

# ❌ 模糊
"帮我做个网站"

# ✅ 清晰
"帮我创建一个电商网站的前端,使用 React + Tailwind CSS,需要包含首页、商品列表、购物车、结账页面,要求响应式设计,支持移动端"

3. 迭代优化

  • 第一次输出后提供反馈
  • 让 Agent 调整和改进
  • 必要时切换 Agent 获取不同视角

4. 质量保证

  • 使用测试部 Agent 进行验证
  • 关键项目使用现实检查员最终审核
  • 重要决策使用多个 Agent 交叉验证

🔧 故障排除

常见问题

Q: Agent 输出不符合预期? A: 提供更详细的任务描述,包括具体要求、技术栈、示例等

Q: 多 Agent 协作效率低? A: 使用编排器自动调度,或手动指定 Agent 顺序

Q: 需要特定领域专业知识? A: 选择对应部门的 Agent,或联系支持获取定制建议

📞 技术支持

📄 许可证

MIT License - 基于 agency-agents 项目改造

Files

19 total
Select a file
Select a file to preview.

Comments

Loading comments…