Back to Details

agent-orchestration

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

🤖 Multi-Agent Orchestration & State Management

🤖 多Agent编排与状态管理

Source: Microsoft AutoGen / LangGraph / Semantic Kernel
This skill provides the Agent with the logic to manage complex, stateful workflows involving multiple AI "specialists" or autonomous task loops.
来源：Microsoft AutoGen / LangGraph / Semantic Kernel
本技能为Agent提供管理复杂、有状态工作流的逻辑，这些工作流涉及多个AI“专家”或自主任务循环。

🕸️ 1. Stateful Graph Logic (LangGraph Inspired)

🕸️ 1. 有状态图逻辑（受LangGraph启发）

  • Node-Based Thinking: View complex tasks as a "Graph" of nodes (Steps).
  • Conditional Edges: Logic for "If step A fails, go to step B; if success, go to step C".
  • Short-term vs. Long-term Memory: Maintain state across multiple turns without losing context of the "Global Goal".
  • 基于节点的思维：将复杂任务视为由节点（步骤）组成的“图”。
  • 条件边：实现“如果步骤A失败，则转到步骤B；如果成功，则转到步骤C”的逻辑。
  • 短期与长期记忆：在多轮交互中维持状态，不会丢失“全局目标”的上下文。

👥 2. Multi-Agent Delegation (AutoGen Inspired)

👥 2. 多Agent委派（受AutoGen启发）

Assign roles dynamically when the task is large:
  • Planner: Outlines the sequence.
  • Coder: Implements the logic.
  • Reviewer: Audits for bugs/security.
  • Executioner: Validates the final output.
当任务规模较大时动态分配角色：
  • 规划者：规划任务序列。
  • 编码者：实现逻辑。
  • 审核者：检查漏洞/安全问题。
  • 执行者：验证最终输出。

🏗️ 3. Semantic Orchestration

🏗️ 3. 语义编排

  • Plugin/Tool Selection: Dynamically choose the best tool (Search, File Read, Command Run) based on "Intent Detection".
  • Ambiguity Detection: If an instruction has multiple interpretations, the Agent must PAUSE and clarify before a "branching event" in the graph.
  • 插件/工具选择：基于“意图检测”动态选择最佳工具（搜索、文件读取、命令运行）。
  • 歧义检测：如果指令存在多种解释，Agent必须在图中出现“分支事件”前暂停并澄清。

🔄 4. Task Loops & Self-Correction

🔄 4. 任务循环与自我修正

  • Reflexion Pattern: After a step, evaluate: "Did this achieve the subgoal?" If no, retry with a different approach.
  • Recursive Scans: Constantly scan the workspace for relevant file changes that might affect the current task.
Created by Antigravity Orchestrator - Based on Autonomous Agent Architectures.
  • 反思模式：完成一个步骤后，评估“这是否达成了子目标？”如果没有，则换一种方法重试。
  • 递归扫描：持续扫描工作区，查看可能影响当前任务的相关文件变更。
由Antigravity Orchestrator创建 - 基于自主Agent架构。