🧱 系统边界

# Agent 边界设计：清晰的职责划分 # Bad: 边界模糊——每个Agent什么都能做 class OmnipotentAgent: """一个Agent拥有所有工具——边界模糊""" tools = ["search", "code", "deploy", "review", "email", "database"] # 问题：工具选择困难、上下文膨胀、职责不清 # Good: 按能力划分边界 class ResearchAgent: """专注信息获取和分析""" tools = ["web_search", "web_fetch", "read_file", "summarize"] boundary = "只负责信息收集和分析，不做执行" class CodeAgent: """专注代码编写和测试""" tools = ["read_file", "write_file", "exec", "test_runner"] boundary = "只负责代码变更，不做部署" class DevOpsAgent: """专注部署和运维""" tools = ["deploy", "monitor", "rollback", "scale"] boundary = "只负责运行时操作，不改变代码" # 边界之间的通信协议 class AgentProtocol: """Agent 之间的通信契约 = API 契约""" def research_to_code(self, research_result): """Research → Code 的接口""" return { "spec": research_result.summary, # 需求规格 "references": research_result.sources, # 参考资料 "constraints": research_result.constraints # 约束条件 } def code_to_devops(self, code_result): """Code → DevOps 的接口""" return { "artifact": code_result.build_path, # 部署物 "config": code_result.config, # 配置 "health_check": code_result.health_url # 健康检查 }