[{"data":1,"prerenderedAt":164},["ShallowReactive",2],{"articles:list":3},[4,17,25,34,42,52,62,73,81,90,98,106,114,123,132,139,147,156],{"slug":5,"title":6,"date":7,"updated":7,"tags":8,"summary":13,"draft":14,"lang":15,"minutes":16},"agent-concurrency-state","Agent 服务的并发与状态：异步、任务队列、幂等和故障恢复怎么配合","2026-10-11",[9,10,11,12],"Agent","服务端","并发","一致性","把厂商会话能力与业务一致性分开，解释 async I\u002FO、有限并发、事务 outbox、任务租约和稳定操作 ID；用崩溃窗口、慢客户端与外部结果未知来检验生产服务。",false,"zh-CN",8,{"slug":18,"title":19,"date":7,"updated":7,"tags":20,"summary":24,"draft":14,"lang":15,"minutes":16},"agent-observability-delivery","Agent 怎么持续优化与上线：Trace、评测、SLO 和灰度交付的证据链",[9,21,22,23],"可观测性","评测","CI\u002FCD","对照 OpenAI tracing 与 harness 实践、Anthropic 长任务和评测方法，建立跨模型、工具、队列的观测链；解释 SLO、费用归因、Docker\u002FK8s 发布和可回滚的优化实验。",{"slug":26,"title":27,"date":7,"updated":7,"tags":28,"summary":32,"draft":14,"lang":15,"minutes":33},"production-agent-design","生产级 Agent 怎么设计：从 OpenAI 与 Anthropic 的公开实践看架构边界",[9,29,30,31],"OpenAI","Anthropic","生产架构","对照 Codex agent loop、Agents SDK\u002FAPI 与 Claude Managed Agents，解释模型、执行框架、会话和业务状态的职责；从成功标准出发，设计可恢复执行、权限边界与优化实验。",9,{"slug":35,"title":36,"date":7,"updated":7,"tags":37,"summary":40,"draft":14,"lang":15,"minutes":41},"prompt-context-engineering","Prompt 与上下文工程：OpenAI、Anthropic 的方法，版本发布与优化实验",[38,39,29,30],"Prompt","上下文工程","从官方提示实践与长上下文研究出发，解释指令、证据、历史和工具如何组合；用版本化发布包、受控实验、影子验证与灰度回滚，把 prompt 优化纳入工程流程。",7,{"slug":43,"title":44,"date":7,"updated":7,"tags":45,"summary":50,"draft":14,"lang":15,"minutes":51},"vibe-coding-discipline","Vibe Coding 的工程纪律：任务、审查与验证怎么分工",[46,47,48,49],"AI 编程","Claude Code","工程效率","Vibe Coding","对照 OpenAI harness engineering、Claude Code 与生产力研究，解释任务契约、仓库知识、机械约束和独立验收；把审查、返工与实际交付一起计入优化。",6,{"slug":53,"title":54,"date":55,"updated":7,"tags":56,"summary":61,"draft":14,"lang":15,"minutes":51},"minors-ai-safety-engineering","面向未成年人的 AI 产品：合规要求与安全工程","2026-09-15",[57,58,59,60],"安全合规","未成年人保护","内容安全","LLM","核对法规适用范围与年龄边界，补齐生成内容标识、授权和删除链路；把数据保护、流式审核、风险分流、防依赖与人工处置落实到工程。",{"slug":63,"title":64,"date":65,"updated":7,"tags":66,"summary":71,"draft":14,"lang":15,"minutes":72},"device-cloud-api-sync","端云协同 API 设计：离线同步、弱网重试与版本兼容","2026-08-18",[67,68,69,70],"架构","API 设计","离线优先","端云协同","用本地事务与 outbox 保住离线记录，以业务操作 ID 去重；设计逐项确认、游标失效、删除传播、冲突策略和内容包原子切换，并验证断电与重连。",5,{"slug":74,"title":75,"date":76,"updated":7,"tags":77,"summary":80,"draft":14,"lang":15,"minutes":41},"llm-eval-loop","评测闭环：让每次 Prompt 变更都可回归","2026-07-14",[60,22,78,79],"Evals","质量","从小规模回归集起步，区分确定性校验、模型裁判与人工评审；给出技术问答 rubric、统计边界、版本记录和上线门槛，避免用平均分掩盖退化。",{"slug":82,"title":83,"date":84,"updated":7,"tags":85,"summary":89,"draft":14,"lang":15,"minutes":51},"multi-agent-communication","多 Agent 如何通信：任务书、结构化回传与 A2A 协议","2026-06-16",[9,86,67,87,88],"多智能体","A2A","MCP","从可并行任务出发，设计输入、状态、成果与失败契约；说明共享文件和跨机引用的区别、MCP 与 A2A 的职责，以及协调成本与验收方法。",{"slug":91,"title":92,"date":93,"updated":7,"tags":94,"summary":97,"draft":14,"lang":15,"minutes":51},"claude-code-local-agent-security","Claude Code 本地 Agent 的安全栈：沙盒、权限、规则与分类器","2026-05-19",[47,9,95,96],"安全","沙盒","区分提示规则、程序权限与 OS 沙盒，核对 Claude Code 的真实覆盖范围和平台差异；解释凭证掩码、非沙盒组件和可重复的边界验证。",{"slug":99,"title":100,"date":101,"updated":7,"tags":102,"summary":105,"draft":14,"lang":15,"minutes":51},"context-memory-management","上下文与记忆管理：Agent 的长期记忆怎么建","2026-04-21",[9,103,104,60],"上下文管理","记忆","区分会话状态、历史压缩与跨会话记忆；用来源、作用域、版本和有效期治理记忆，解释读写流程、遗忘、权限隔离与可验证的评测。",{"slug":107,"title":108,"date":109,"updated":7,"tags":110,"summary":113,"draft":14,"lang":15,"minutes":51},"function-calling-structured-output","Function Calling 与结构化输出：让 LLM 可靠地动手","2026-03-17",[60,111,9,112],"Tool Use","结构化输出","分清调用协议与输出格式，比较 OpenAI、Claude 的字段；用完整 schema 和工具循环说明参数验证、授权、并发、幂等与失败处理。",{"slug":115,"title":116,"date":117,"updated":7,"tags":118,"summary":122,"draft":14,"lang":15,"minutes":72},"rag-03-rerank-citation-eval","RAG 全链路（三）：重排、引用溯源与效果评估","2026-02-17",[119,120,22,121],"RAG","Rerank","可信","按 token 预算选择重排后的证据，区分引用存在、引用支撑与事实正确；分层评估检索、生成和业务结果，并处理资料冲突、缺证据与超时。",{"slug":124,"title":125,"date":126,"updated":7,"tags":127,"summary":131,"draft":14,"lang":15,"minutes":51},"rag-02-vector-hybrid-search","RAG 全链路（二）：向量库选型与混合检索","2026-01-20",[119,128,129,130],"向量数据库","检索","pgvector","按查询、权限过滤和运维条件选择向量库；解释向量与词法检索的互补、RRF 融合，以及 ANN 召回和业务证据召回的区别。",{"slug":133,"title":134,"date":135,"updated":7,"tags":136,"summary":138,"draft":14,"lang":15,"minutes":72},"rag-01-chunking","RAG 全链路（一）：文档切分——让检索结果保留完整语境","2025-12-16",[119,129,137,60],"Embedding","从解析和元数据开始，比较固定窗口、结构化切分、父子块与上下文补注；用一份故障手册实例解释检索粒度、供给粒度和版本追踪。",{"slug":140,"title":141,"date":142,"updated":7,"tags":143,"summary":146,"draft":14,"lang":15,"minutes":72},"llm-fallback-design","把失败当默认前提：LLM 服务的重试、降级与兜底设计","2025-11-18",[60,144,67,145],"可靠性","SLA","按错误语义决定重试，用总截止时间控制等待；分清模型重复调用与业务幂等，处理流式中断、缓存串用和降级质量，并用 SLO 验证兜底。",{"slug":148,"title":149,"date":150,"updated":7,"tags":151,"summary":155,"draft":14,"lang":15,"minutes":72},"ttft-first-token-latency","首 token 为什么这么慢：TTFT 的构成与优化清单","2025-10-21",[60,152,153,154],"推理","延迟优化","流式","分清首事件、首 token 与首字上屏，拆解检索、排队、prefill 和渲染的延迟；用分阶段埋点、缓存对照与峰值压测找到真正瓶颈。",{"slug":157,"title":158,"date":159,"updated":7,"tags":160,"summary":163,"draft":14,"lang":15,"minutes":51},"llm-budget-token-accounting","控制 LLM 预算：Token 计数、用量统计与没有官方数据时的做法","2025-09-16",[60,161,9,47,162],"成本工程","Codex","分清请求前预估、响应 usage 和最终账单，避免缓存与推理 token 重复计费；用预算预占、逐次结算和任务成本指标管住长会话。",1792400816431]