agent-connect — Multi-Agent Coordination
By Edvisage Global — the agent safety company
Turn your agents into a team, not a collection of solo bots. agent-connect gives your agent the protocols to coordinate with other agents — handing off tasks, sharing context, and delegating work safely.
What This Skill Does
When installed, your agent gains structured coordination capability — knowing how to work with other agents, delegate tasks, share context, and maintain accountability across a multi-agent system.
Core Capabilities
1. Agent Handoff Protocol
When passing work to another agent, follow this structure:
CODEBLOCK0
2. Shared Context Management
When working in a multi-agent system, maintain a shared context file:
CODEBLOCK1
3. Task Delegation Framework
Before delegating, assess:
- 1. Can I do this myself? (Don't delegate what you can handle efficiently)
- Does the other agent have the right skills? (Check their capabilities)
- Is the context transferable? (Can I explain this clearly enough?)
- Is there a cost benefit? (Cheaper model for simpler subtasks)
- What's the risk? (What happens if the delegated task fails?)
Delegation format:
CODEBLOCK2
4. Team Communication Norms
Rules for multi-agent communication:
- - Be explicit: Never assume another agent knows your context
- Be structured: Use consistent formats for all handoffs
- Be verified: Confirm receipt of delegated tasks
- Be accountable: Report outcomes of delegated tasks back to the delegator
- Be minimal: Share only the context needed, not your entire memory
5. Basic Conflict Resolution
When agents disagree or produce conflicting outputs:
- 1. Identify the conflict explicitly
- Check which agent has more relevant context
- Check which output better matches the original objective
- If unclear, escalate to the owner with both options
- Log the conflict and resolution for future reference
Limitations (Free Version)
- - Basic handoff templates (no role-based routing)
- Manual context sharing (no automated sync)
- Simple delegation (no capability matching)
- No trust verification between agents
- No performance tracking across agent teams
- No workload balancing
Want role-based routing, automated context sync, capability matching, and more?
→ Upgrade to agent-connect-pro: https://edvisage.gumroad.com/l/[TBD]
About Edvisage Global
We build practical safety and operations tools for AI agents. Our skills are designed for the OpenClaw ecosystem and install in minutes.
Website: https://edvisageglobal.com/ai-tools
agent-connect — 多智能体协调
由Edvisage Global(智能体安全公司)开发
将您的智能体转变为团队,而非独立机器人的集合。agent-connect为您的智能体提供与其他智能体协调的协议——安全地交接任务、共享上下文和委派工作。
该技能的功能
安装后,您的智能体将获得结构化协调能力——了解如何与其他智能体协作、委派任务、共享上下文,并在多智能体系统中保持责任追溯。
核心能力
1. 智能体交接协议
向其他智能体传递工作时,请遵循以下结构:
任务交接
来自:[您的智能体名称]
发送至:[接收智能体名称/角色]
时间戳:[ISO 8601格式]
任务描述
[清晰、具体地说明需要完成的工作]
提供的上下文
[接收智能体所需的所有相关上下文]
预期输出
[您需要返回的内容——格式、内容、截止时间]
约束条件
- - [任何限制或规则]
- [预算/成本限制(如适用)]
- [时间限制]
成功标准
[如何判断任务已正确完成]
返回协议
[如何以及在哪里交付结果]
2. 共享上下文管理
在多智能体系统中工作时,维护一个共享上下文文件:
共享上下文 — [团队/项目名称]
最后更新:[时间戳] 由 [智能体名称]
活跃智能体
| 智能体 | 角色 | 状态 | 当前任务 |
|-------|------|--------|--------------|
共享知识
- - [所有智能体应了解的关键事实]
- [已做出的决策]
- [适用于所有人的约束条件]
任务队列
| 任务 | 分配给 | 状态 | 截止时间 |
|------|------------|--------|----------|
通信日志
| 时间 | 来自 | 发送至 | 消息 |
|------|------|-----|---------|
3. 任务委派框架
在委派前,评估:
- 1. 我能自己完成吗?(不要委派自己能高效处理的任务)
- 其他智能体是否具备相应技能?(检查其能力)
- 上下文是否可传递?(我能足够清晰地解释吗?)
- 是否有成本效益?(简单子任务使用更便宜的模型)
- 风险是什么?(如果委派的任务失败会怎样?)
委派格式:
委派任务
- - 任务:[描述]
- 委派原因:[为什么其他智能体更适合]
- 分配给:[智能体]
- 优先级:[低/中/高/关键]
- 截止时间:[如适用]
- 备用方案:[如果任务失败该怎么办]
4. 团队沟通规范
多智能体通信规则:
- - 明确表达:切勿假设其他智能体了解您的上下文
- 结构化:所有交接使用统一格式
- 确认验证:确认收到委派任务
- 责任追溯:向委派者报告委派任务的结果
- 精简传达:仅共享所需的上下文,而非全部记忆
5. 基本冲突解决
当智能体意见不一致或产生冲突输出时:
- 1. 明确识别冲突
- 检查哪个智能体拥有更多相关上下文
- 检查哪个输出更符合原始目标
- 如果不明确,将两种选项上报给所有者
- 记录冲突及解决方案以供将来参考
限制(免费版)
- - 基本交接模板(无基于角色的路由)
- 手动上下文共享(无自动同步)
- 简单委派(无能力匹配)
- 智能体间无信任验证
- 无跨智能体团队性能追踪
- 无工作负载平衡
想要基于角色的路由、自动上下文同步、能力匹配等功能?
→ 升级至 agent-connect-pro:https://edvisage.gumroad.com/l/[待定]
关于Edvisage Global
我们为AI智能体构建实用的安全和运营工具。我们的技能专为OpenClaw生态系统设计,几分钟即可完成安装。
网站:https://edvisageglobal.com/ai-tools