一键导入
a2a-communication
Procedural knowledge for orchestrating multi-agent systems and Agent-to-Agent (A2A) handoffs.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Procedural knowledge for orchestrating multi-agent systems and Agent-to-Agent (A2A) handoffs.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Procedural knowledge for generating and maintaining code documentation, API specs, and architecture diagrams.
Procedural knowledge for designing autonomous AI agents with tools, memory, and reasoning loops.
Procedural knowledge for implementing the Model Context Protocol (MCP) to expose local tools to LLMs.
Procedural knowledge for ensuring frontend interfaces comply with WCAG standards.
Procedural knowledge for designing and configuring continuous integration and deployment pipelines.
Procedural knowledge for standardizing error boundaries, logging, and graceful degradation.
| name | a2a-communication |
| description | Procedural knowledge for orchestrating multi-agent systems and Agent-to-Agent (A2A) handoffs. |
Procedural knowledge for orchestrating multi-agent systems and Agent-to-Agent (A2A) handoffs.
Enable the AI Expert and Architect roles to design scalable multi-agent ecosystems where specialized agents can securely delegate tasks, share context, and resolve conflicts.
Step 1: Define the Topology
- Supervisor/Hierarchical: A router agent delegates tasks to worker agents and aggregates results.
- Sequential/Chain: Agent A finishes its task and passes the output directly to Agent B.
- Blackboard/Shared State: All agents read from and write to a centralized shared state.
Step 2: Context Handoff
- Define the exact schema for the handoff object.
- Include: Objective, Current State, Completed Steps, and Constraints.
- Do NOT pass the entire raw chat history to the next agent; pass a structured summary to save tokens.
Step 1: Delegation
- Give the Supervisor agent a `delegate_task(agent_id, instructions)` tool.
- Ensure the worker agent has a mechanism to signal `task_complete(result)`.
Step 2: Conflict Resolution
- If the Reviewer agent rejects the Writer agent's work, the Reviewer must provide structured feedback (`<feedback>`) and transition state back to the Writer.
- Implement a maximum recursion depth to prevent endless debate between agents.
"Apply the a2a-communication skill from .ace/skills/a2a-communication/SKILL.md
to design the routing logic for the Researcher and Writer agents."