<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0"
     xmlns:atom="http://www.w3.org/2005/Atom"
     xmlns:content="http://purl.org/rss/1.0/modules/content/"
     xmlns:dc="http://purl.org/dc/elements/1.1/"
     xmlns:media="http://search.yahoo.com/mrss/">
  <channel>
    <title>AI Agent on 熊鑫伟（cubxxw）· 公开的实践与思想档案</title>
    <link>https://cubxxw.com/zh/ai-agent/</link>
    <description>熊鑫伟的公开实践与思想档案，记录 AI 工作流、真实项目、工程方法、系统思维，以及关于个人成长、哲学与人生的长期思考。</description>
    <image>
      <title>熊鑫伟（cubxxw）· 公开的实践与思想档案</title>
      <url>https://cubxxw.com/assets/og-image.png</url>
      <link>https://cubxxw.com/</link>
    </image>
    <generator>Hugo -- gohugo.io</generator>
    <language>zh-cn</language>
    <lastBuildDate>Fri, 07 Aug 2026 19:43:00 +0800</lastBuildDate>
    <atom:link href="https://cubxxw.com/zh/ai-agent/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Agent 不是一种产品：十套系统如何重新分配控制权、状态、身份与副作用</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/agent-system-design-synthesis/</link>
      <pubDate>Fri, 07 Aug 2026 19:43:00 +0800</pubDate>
      <atom:updated>Fri, 07 Aug 2026 19:43:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/agent-system-design-synthesis/</guid>
      <description>基于十篇个案，以控制权、状态、执行身体、强制层和产品表面五个所有权问题，推导 bounded function、workflow、harness、event platform、gateway/computer 与 vertical organization 六种形态，并给出副作用宪法、五层停止协议和选型树。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">System Design</category>
      <category domain="tag">Data Processing</category>
      <category domain="tag">Security</category>
      <category domain="tag">Development</category>
      <enclosure url="https://cubxxw.com/images/agent-system-series/11-synthesis/agent-system-constitutional-route.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/agent-system-series/11-synthesis/agent-system-constitutional-route.svg" medium="image"><media:description>Agent 系统设计的五个所有权问题、从意图到副作用与证据的控制权宪法之河，以及六种最小成立形态的决策图</media:description></media:content>
    </item>
    <item>
      <title>Conversation as Database：OpenHands 的无状态 Agent 与事件运行时</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/agent-system-design-openhands/</link>
      <pubDate>Fri, 07 Aug 2026 19:40:00 +0800</pubDate>
      <atom:updated>Fri, 07 Aug 2026 19:40:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/agent-system-design-openhands/</guid>
      <description>以 OpenHands SDK v1.41.0 为样本，沿事件链拆解无状态 step、Conversation 与 local/remote workspace。文章区分 event log、覆盖式 base state、派生 view 和外部世界，并以崩溃路径说明内部可回放为何仍不等于副作用恰好一次。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Data Processing</category>
      <category domain="tag">Monitoring</category>
      <category domain="tag">Security</category>
      <category domain="tag">Development</category>
      <enclosure url="https://cubxxw.com/images/agent-system-series/10-openhands/openhands-event-sourced-runtime.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/agent-system-series/10-openhands/openhands-event-sourced-runtime.svg" medium="image"><media:description>OpenHands V1 从无状态 Agent step、追加式 typed event log、base state、派生 view 到 local/remote runtime 与副作用恢复协议的架构图</media:description></media:content>
    </item>
    <item>
      <title>多 Agent 辩论真的增加了信息吗：TradingAgents 的组织图与相关性风险</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/agent-system-design-tradingagents/</link>
      <pubDate>Fri, 07 Aug 2026 18:50:00 +0800</pubDate>
      <atom:updated>Fri, 07 Aug 2026 18:50:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/agent-system-design-tradingagents/</guid>
      <description>以 TradingAgents v0.3.1 与论文 v7 为样本，把分析师、牛熊辩手、交易员、风险声部和组合经理还原成真实 LangGraph。文章区分角色多样性与证据独立性，核验串行分析、共享模型、时间截点、checkpoint 和反思记忆，并以相关性、消融及外部执行成本检验多 Agent 辩论是否真正增加信息。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Data Processing</category>
      <category domain="tag">Monitoring</category>
      <category domain="tag">Security</category>
      <category domain="tag">Development</category>
      <enclosure url="https://cubxxw.com/images/agent-system-series/09-tradingagents/tradingagents-evidence-debate-graph.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/agent-system-series/09-tradingagents/tradingagents-evidence-debate-graph.svg" medium="image"><media:description>TradingAgents 从时间截点证据、串行分析师、牛熊法庭、风险声部到组合经理研究输出的证据法庭架构图</media:description></media:content>
    </item>
    <item>
      <title>Instructions as Code：OpenMontage 如何用工件契约编排视频生产</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/agent-system-design-openmontage/</link>
      <pubDate>Fri, 07 Aug 2026 18:20:00 +0800</pubDate>
      <atom:updated>Fri, 07 Aug 2026 18:20:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/agent-system-design-openmontage/</guid>
      <description>以 OpenMontage 固定提交为样本，沿脚本、分镜、素材、剪辑、渲染与导出追踪规范工件，解释 coding assistant 如何用 YAML 和 Skill 驱动生产。文章拆开 tool、checkpoint、validator 与人工门，并以无音轨成片和付费素材崩溃检验恢复、成本、版权及发布边界。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Automation</category>
      <category domain="tag">Content Strategy</category>
      <category domain="tag">Development</category>
      <category domain="tag">Open Source</category>
      <enclosure url="https://cubxxw.com/images/agent-system-series/08-openmontage/openmontage-artifact-production-line.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/agent-system-series/08-openmontage/openmontage-artifact-production-line.svg" medium="image"><media:description>OpenMontage 由 instruction 控制带、canonical artifact 胶片河和工具检查点验证带组成的视频生产架构图</media:description></media:content>
    </item>
    <item>
      <title>拒绝成为 Agent：TaxHacker 为什么把 LLM 压缩成一个函数</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/agent-system-design-taxhacker/</link>
      <pubDate>Fri, 07 Aug 2026 17:55:00 +0800</pubDate>
      <atom:updated>Fri, 07 Aug 2026 17:55:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/agent-system-design-taxhacker/</guid>
      <description>以 TaxHacker v0.8.5 为样本，追踪票据如何经过多模态抽取、provider fallback、动态 schema、表单校验、重复检查与人工保存，解释何时应把 LLM 限制成结构化函数。文章区分结构有效、事实正确与会计记录，并揭示 compatible 路径、前四页预览、缓存候选和金融数据的边界。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Document AI</category>
      <category domain="tag">Data Processing</category>
      <category domain="tag">Security</category>
      <category domain="tag">Development</category>
      <enclosure url="https://cubxxw.com/images/agent-system-series/07-taxhacker/taxhacker-bounded-ai.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/agent-system-series/07-taxhacker/taxhacker-bounded-ai.svg" medium="image"><media:description>TaxHacker 将不可信票据经顺序 provider 抽取、schema 窄腰、原件核对、表单校验与显式保存变成 canonical transaction 的架构图</media:description></media:content>
    </item>
    <item>
      <title>OpenClaw 的常驻网关：连续性不是把所有消息塞进同一会话</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/agent-system-design-openclaw/</link>
      <pubDate>Fri, 07 Aug 2026 17:40:00 +0800</pubDate>
      <atom:updated>Fri, 07 Aug 2026 17:40:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/agent-system-design-openclaw/</guid>
      <description>以 OpenClaw v2026.7.1-2 为样本，追踪消息如何经过渠道准入、binding、agent 与 session key，解释跨渠道连续性的身份维度。文章拆开 Gateway、模型、workspace、sandbox 与设备节点，并用跨账户串线反例说明常驻 Agent 的权限、凭据、注入与撤销边界。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Security</category>
      <category domain="tag">Monitoring</category>
      <category domain="tag">Automation</category>
      <category domain="tag">Development</category>
      <enclosure url="https://cubxxw.com/images/agent-system-series/06-openclaw/openclaw-persistent-gateway.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/agent-system-series/06-openclaw/openclaw-persistent-gateway.svg" medium="image"><media:description>OpenClaw 从互联网渠道经过常驻 Gateway、确定性 bindings、session key、per-agent enclave 到设备节点与外部副作用的路由图</media:description></media:content>
    </item>
    <item>
      <title>n8n 的确定性外骨骼：Queue、Worker 与副作用从不承诺 Exactly-Once</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/agent-system-design-n8n/</link>
      <pubDate>Fri, 07 Aug 2026 17:25:00 +0800</pubDate>
      <atom:updated>Fri, 07 Aug 2026 17:25:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/agent-system-design-n8n/</guid>
      <description>以 n8n 2.33.6 源码追踪一次执行如何跨主进程、Redis 队列、工作进程与 Postgres 流动，并拆开工作流图与 Agent 节点的控制边界。文章解释等待、重试、崩溃恢复、幂等键和对账各自解决什么，以及 queue mode 为什么只能扩大吞吐，不能承诺 exactly-once 副作用。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Automation</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Monitoring</category>
      <category domain="tag">Security</category>
      <category domain="tag">Development</category>
      <enclosure url="https://cubxxw.com/images/agent-system-series/05-n8n/n8n-deterministic-spine.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/agent-system-series/05-n8n/n8n-deterministic-spine.svg" medium="image"><media:description>n8n 确定性 workflow 主干、受限 Agent 支线、人工与幂等闸门、外部回执对账以及 Redis worker Postgres queue-mode 车场</media:description></media:content>
    </item>
    <item>
      <title>给 Agent 一台电脑之后：Manus 如何把回复变成交付物</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/agent-system-design-manus/</link>
      <pubDate>Fri, 07 Aug 2026 17:15:00 +0800</pubDate>
      <atom:updated>Fri, 07 Aug 2026 17:15:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/agent-system-design-manus/</guid>
      <description>基于 Manus 2026-08-07 第一方资料，拆开临时 Sandbox、持久 Cloud Computer、Cloud Browser、Browser Operator 与 Wide Research。文章解释计算机如何把回复升级为可验收工件和持续服务，也说明登录态、外部副作用与恢复责任为何仍在 VM 边界外。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Product Strategy</category>
      <category domain="tag">Security</category>
      <category domain="tag">Automation</category>
      <category domain="tag">Cloud</category>
      <enclosure url="https://cubxxw.com/images/agent-system-series/04-manus/manus-computer-to-artifact.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/agent-system-series/04-manus/manus-computer-to-artifact.svg" medium="image"><media:description>Manus task envelope、临时 Sandbox VM、browser files code software 流水线、artifact 验收、Wide Research 与本地 Browser Operator 边界</media:description></media:content>
    </item>
    <item>
      <title>Codex 的协议化内核：第二个 Surface 不该复制第一个 Agent</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/agent-system-design-codex/</link>
      <pubDate>Fri, 07 Aug 2026 17:05:00 +0800</pubDate>
      <atom:updated>Fri, 07 Aug 2026 17:05:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/agent-system-design-codex/</guid>
      <description>以 Codex 0.147.0、App Server 契约和 Rust core 为证据，拆开 Thread、Turn、Item 与 Submission、Event 两层协议。文章解释多种客户端如何共享 Agent loop、状态与治理，并厘清 approval、sandbox、持久化与产品 Surface 的边界。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Open Source</category>
      <category domain="tag">Development</category>
      <category domain="tag">Security</category>
      <category domain="tag">MCP</category>
      <enclosure url="https://cubxxw.com/images/agent-system-series/03-codex/codex-event-control-plane.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/agent-system-series/03-codex/codex-event-control-plane.svg" medium="image"><media:description>Codex 多 Surface、App Server 公共契约、Submission 与 Event 双总线以及 approval 和 sandbox 副作用闸门</media:description></media:content>
    </item>
    <item>
      <title>Pi 的减法：最小 Agent Kernel 保留什么，责任又去了哪里</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/agent-system-design-pi/</link>
      <pubDate>Fri, 07 Aug 2026 16:40:00 +0800</pubDate>
      <atom:updated>Fri, 07 Aug 2026 16:40:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/agent-system-design-pi/</guid>
      <description>以 Pi v0.84.1 源码为证据，沿 agent loop、默认四工具、JSONL session tree 与 extension seam 反推最小 Agent kernel。文章审计 Pi 省略 Plan、MCP、subagent、permission 和 sandbox 后，安全、一致性和恢复责任的去向。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Open Source</category>
      <category domain="tag">Development</category>
      <category domain="tag">Security</category>
      <category domain="tag">Context Engineering</category>
      <enclosure url="https://cubxxw.com/images/agent-system-series/02-pi/pi-minimal-kernel.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/agent-system-series/02-pi/pi-minimal-kernel.svg" medium="image"><media:description>Pi 最小 Agent kernel、可拆卸 extension 轨道、provider 与 host OS 信任边界以及下方 JSONL session tree</media:description></media:content>
    </item>
    <item>
      <title>Claude Code 的扩展语法：一条仓库修改怎样逐级获得控制</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/agent-system-design-claude-code/</link>
      <pubDate>Fri, 07 Aug 2026 16:35:00 +0800</pubDate>
      <atom:updated>Fri, 07 Aug 2026 16:35:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/agent-system-design-claude-code/</guid>
      <description>以 Claude Code v2.1.224 为证据边界，沿仓库修改控制链拆解 CLAUDE.md、Skill、MCP、Hook、permission、sandbox、subagent 与 worktree。文章区分软语境、能力接口、确定性拦截和文件隔离，说明模型何时拥有下一步，代码、操作系统、测试与人在何处接管后果。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Context Engineering</category>
      <category domain="tag">MCP</category>
      <category domain="tag">Security</category>
      <category domain="tag">Testing</category>
      <enclosure url="https://cubxxw.com/images/agent-system-series/01-claude-code/claude-code-progressive-control.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/agent-system-series/01-claude-code/claude-code-progressive-control.svg" medium="image"><media:description>Claude Code 中模糊意图经过上下文、Agent loop、工具提案、权限、沙箱、工作树和测试，成为受约束仓库修改的控制河流</media:description></media:content>
    </item>
    <item>
      <title>Claude Tag 深度拆解：Slack 里的共享 AI，正在长成组织级 Agent 运行时</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/claude-tag-organizational-agent-runtime/</link>
      <pubDate>Wed, 05 Aug 2026 00:00:00 +0800</pubDate>
      <atom:updated>Wed, 05 Aug 2026 00:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/claude-tag-organizational-agent-runtime/</guid>
      <description>Claude Tag 把共享 AI 带进 Slack，把 Agent 的身份、权限、组织记忆与主动执行放进同一套系统。本文从线程级沙箱、Agent Proxy、频道权限和异步任务拆解其技术架构，分析产品价值、用户体验与治理风险，对比钉钉 AI 小钉、DWS 和飞书 Aily，并预测企业协作 Agent 的下一阶段形态。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Automation</category>
      <category domain="tag">Context Engineering</category>
      <category domain="tag">Product Strategy</category>
      <category domain="tag">Security</category>
      <category domain="tag">Productivity</category>
      <category domain="tag">LLM</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/claude-tag-organizational-agent-runtime.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/claude-tag-organizational-agent-runtime.jpeg" medium="image"><media:description>多人共享的组织级 Agent 调度台、权限柜、记忆库与隔离执行空间</media:description></media:content>
    </item>
    <item>
      <title>我知道什么是 Agent，却说不出它的定义</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/2026-08-02-ai-era-conceptual-mobility/</link>
      <pubDate>Sun, 02 Aug 2026 18:28:00 +0800</pubDate>
      <atom:updated>Sun, 02 Aug 2026 18:28:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/2026-08-02-ai-era-conceptual-mobility/</guid>
      <description>我能辨认 Agent 的边界，却无法立刻给出一句没有漏洞的定义。这篇文章从一次真实的交流卡顿出发，提出一套概念转换方法：把体感拆成正反例和区分轴，放回关系模型，按行动后果压缩为工作定义，再写成可执行、可失败的 eval，并让现实反馈持续修正它。它适合正在讨论 Agent、知识与经验，却总在同一个词上错位的人。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Context Engineering</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Philosophy</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/2026-08-02-ai-era-conceptual-mobility.webp" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/2026-08-02-ai-era-conceptual-mobility.webp" medium="image"><media:description>散乱木片依次穿过三道透明取景框，被分组后越过边界门进入测试棋盘，结果沿回路返回起点</media:description></media:content>
    </item>
    <item>
      <title>常识不是本质：AI 时代如何从需求、约束与稀缺性重构行业</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/common-sense-ai-industry-reconstruction/</link>
      <pubDate>Sat, 01 Aug 2026 23:00:47 +0800</pubDate>
      <atom:updated>Sat, 01 Aug 2026 23:00:47 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/common-sense-ai-industry-reconstruction/</guid>
      <description>这篇文章从常识是不是本质出发，区分常识、共识、本质与第一性原理，再用奈飞从 DVD 到流媒体的迁移检验一套行业分析工序：先找稳定需求，画出供需与利益结构，识别 AI 让哪些旧约束消失、稀缺性如何迁移，最后用招聘案例完整说明怎样把新方案写成可解释、可验收、能由生产失败持续校准的 Agent 工作流全貌。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Product Strategy</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Automation</category>
      <category domain="tag">Career</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/common-sense-ai-industry-reconstruction.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/common-sense-ai-industry-reconstruction.jpeg" medium="image"><media:description>固定需求轴与被重新连接的行业工作流卡片</media:description></media:content>
    </item>
    <item>
      <title>AI 时代，遗忘比记忆更难：一个活系统为什么必须会放下</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/2026-07-31-forgetting-is-an-ai-system-capability/</link>
      <pubDate>Fri, 31 Jul 2026 01:10:00 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 01:10:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/2026-07-31-forgetting-is-an-ai-system-capability/</guid>
      <description>当存储和检索越来越便宜，个人知识库与 Agent 更难解决的是信息过期与身份判断越权。本文通过与认知神经科学家和 AI 记忆工程师的模拟访谈，讨论遗忘为何是一种判断能力，以及怎样为长期运行的智能系统设计可审计的衰减机制。文章区分降温、归档和撤销，并解释哪些记忆可以自治淘汰、哪些必须保留人的最终授权。
</description>
      <category domain="category">Technology</category>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Context Engineering</category>
      <category domain="tag">Self-Discovery</category>
      <category domain="tag">Philosophy</category>
      <category domain="tag">Personal Growth</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/deep-interviews/forgetting-is-an-ai-system-capability.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/deep-interviews/forgetting-is-an-ai-system-capability.jpeg" medium="image"><media:description>分层记忆系统让活跃节点升温、旧信息沉降并保留审计轨迹</media:description></media:content>
    </item>
    <item>
      <title>Open Design 0.16.1：给 Coding Agent 一张可复用的设计工作台</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/open-design/</link>
      <pubDate>Wed, 22 Jul 2026 20:30:00 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 16:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/open-design/</guid>
      <description>本文固定 Open Design 0.16.1，拆解插件、功能技能、设计模板与设计系统四个平面，比较本地代理、Open Design Cloud 与多提供商 BYOK 的数据边界，并厘清运行时目录、模型流量、遥测、SSRF 防护和服务商配额的差别，帮助 Coding Agent 用户判断它能否进入自己的设计工作流。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Project Learning</category>
      <category domain="tag">MCP</category>
      <category domain="tag">Open Source</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/open-design.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/open-design.jpeg" medium="image"><media:description>Open Design 作为 Coding Agent 的四平面设计工作台</media:description></media:content>
    </item>
    <item>
      <title>Claude Code 实战手册：从验证闭环到并行代理的 10 个配置</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/claude-code-boris-121-tips-playbook/</link>
      <pubDate>Mon, 20 Jul 2026 23:30:00 +0800</pubDate>
      <atom:updated>Mon, 20 Jul 2026 23:30:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/claude-code-boris-121-tips-playbook/</guid>
      <description>Claude Code 实战手册把官方能力整理成十个改动：精简 CLAUDE.md、建立验证闭环、按风险配置权限、用 worktree 隔离并行任务，正确选择 plan、auto、goal、loop 与 Routines。每节说明适用边界、套餐限制和验收方法，帮助开发者把偶然的好结果沉淀成可重复、可审查的工程系统。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Automation</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Productivity</category>
      <category domain="tag">Testing</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/claude-code-boris-121-tips-playbook.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/claude-code-boris-121-tips-playbook.jpeg" medium="image"><media:description>Claude Code 十项实战配置：验证、权限、上下文、循环与并行代理</media:description></media:content>
    </item>
    <item>
      <title>独立开发者的 Loop Engineering：验证器、状态与安全自动化</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/prompt-loop-engineering-practice/</link>
      <pubDate>Mon, 20 Jul 2026 21:00:00 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 12:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/prompt-loop-engineering-practice/</guid>
      <description>Loop Engineering 把反复提示变成可验证、可停机的工程系统。本文结合真实仓库实践，拆解状态分层、确定性闸门、评分器、只读验证、权限梯度与无人值守护栏，并厘清 Claude Code /goal 与 Codex Goals 的机制差异，帮助独立开发者在保留判断力的前提下长期安全放大产能。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Automation</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Testing</category>
      <category domain="tag">Solo Builder</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/prompt-loop-engineering-practice.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/prompt-loop-engineering-practice.jpeg" medium="image"><media:description>一张象征验证器、状态与安全自动化的安静控制台</media:description></media:content>
    </item>
    <item>
      <title>独立创作者的 AI 视频剪辑流水线：拍摄、粗剪与多平台发布</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/solo-creator-video-pipeline/</link>
      <pubDate>Sun, 19 Jul 2026 22:00:00 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 12:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/solo-creator-video-pipeline/</guid>
      <description>面向独立创作者的AI视频剪辑流水线：先用镜头清单约束拍摄，再由ffmpeg、faster-whisper、PySceneDetect和结构化EDL完成粗剪，把节奏与判断留给人。文章核验截至2026年7月31日的平台发布边界、AI内容标识和Sora停服时间，帮助开发者降低重复劳动并守住原创表达与账号安全。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">Automation</category>
      <category domain="tag">Super Individual</category>
      <category domain="tag">Content Strategy</category>
      <category domain="tag">Solo Builder</category>
      <category domain="tag">Harness Engineering</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/solo-creator-video-pipeline.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/solo-creator-video-pipeline.jpeg" medium="image"><media:description>独立创作者从拍摄到多平台发布的 AI 视频剪辑流水线</media:description></media:content>
    </item>
    <item>
      <title>当东西人人做得出来，「这是他做的」就成了信号</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/super-individual-stack-reputation/</link>
      <pubDate>Sun, 19 Jul 2026 11:00:00 +0800</pubDate>
      <atom:updated>Sun, 19 Jul 2026 11:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/super-individual-stack-reputation/</guid>
      <description>当建造成本趋近于零，做出一个东西本身不再能证明任何事，人们只好去看是谁做的。这篇讲装备栈的第一层——声誉层：为什么信任是 AI 唯一抹不平的东西，它靠哪三种载体存在，为什么它永远不会出现在任何一份工具清单里，以及它最难堪的三个真相。最后回答整个系列的问题：怎么把生产、判断、分发、声誉四层接成一个互相喂养的闭环，而不是四件互不相干的事。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">Super Individual</category>
      <category domain="tag">Open Source</category>
      <category domain="tag">Solo Builder</category>
      <category domain="tag">Personal Growth</category>
      <category domain="tag">Product Strategy</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/super-individual-stack-reputation.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/super-individual-stack-reputation.jpeg" medium="image"><media:description>当东西人人做得出来，「这是他做的」就成了信号</media:description></media:content>
    </item>
    <item>
      <title>独立开发者的分发策略：产品能被复制，读者关系不能</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/super-individual-stack-distribution/</link>
      <pubDate>Sun, 19 Jul 2026 10:30:00 +0800</pubDate>
      <atom:updated>Sat, 01 Aug 2026 03:35:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/super-individual-stack-distribution/</guid>
      <description>写给独立开发者的分发策略：把产品过程沉淀成可检索、可验证的内容，按渠道功能选择主场和入口，用七天、三十天、九十天的归因闭环衡量从曝光、访问、订阅到激活与结果的路径，同时守住隐私、授权、平台规则和安全边界，让受众关系成为长期可复用的资产，而不是依赖算法的一次流量，并在复盘里分清内容问题、渠道问题、产品问题与留存问题。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">Super Individual</category>
      <category domain="tag">Content Strategy</category>
      <category domain="tag">GEO</category>
      <category domain="tag">Solo Builder</category>
      <category domain="tag">Product Strategy</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/super-individual-stack-distribution.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/super-individual-stack-distribution.jpeg" medium="image"><media:description>独立开发者的分发策略：产品可以复制，读者关系需要长期积累</media:description></media:content>
    </item>
    <item>
      <title>AI 提需求不要钱，这才是最危险的地方</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/super-individual-stack-judgment/</link>
      <pubDate>Sun, 19 Jul 2026 10:00:00 +0800</pubDate>
      <atom:updated>Sun, 19 Jul 2026 10:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/super-individual-stack-judgment/</guid>
      <description>当执行成本趋近于零，做错方向的代价不会变小，只会变成全额。AI 提需求几乎不要成本，队列会被一堆看着合理、实际低价值的任务填满，而 agent 会忠实地全部做完。这篇拆解判断层的三件事：需求闸门怎么守、验收标准怎么写、怎么把判断外化成 agent 每次开工都会读的规则文件——那是这一层唯一能复利的部分。也诚实指出规则装不下的那半判断，以及为什么命中率才是判断力的真实检验标准。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Super Individual</category>
      <category domain="tag">Context Engineering</category>
      <category domain="tag">Product Strategy</category>
      <category domain="tag">Solo Builder</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/super-individual-stack-judgment.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/super-individual-stack-judgment.jpeg" medium="image"><media:description>AI 提需求不要钱，这才是最危险的地方</media:description></media:content>
    </item>
    <item>
      <title>夜里跑十个 agent，早上收十个 PR，然后我卡住了</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/super-individual-stack-production/</link>
      <pubDate>Sun, 19 Jul 2026 09:30:00 +0800</pubDate>
      <atom:updated>Sun, 19 Jul 2026 09:30:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/super-individual-stack-production/</guid>
      <description>《超级个体的装备栈》第二篇，讲最底下那层——生产层。这一层的所有进步都是公共品，同步发放给每个人，所以配到够用就该停手。真正会卡住你的不是执行带宽，而是 review 带宽，它几乎无法用同样的方式扩张。唯一有效的扩张手段是提高一次做对率，把规则前置、验收标准写进工单、让 agent 自己跑到测试全绿再开 PR。后半段给出 AI 代码质量与安全的实证数据，它们正好解释了为什么测试和隔离才是这层真正的装备。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Super Individual</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Testing</category>
      <category domain="tag">Security</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/super-individual-stack-production.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/super-individual-stack-production.jpeg" medium="image"><media:description>夜里跑十个 agent，早上收十个 PR，然后我卡住了</media:description></media:content>
    </item>
    <item>
      <title>超级个体的四层 AI 装备栈：什么才会真正复利</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/super-individual-stack-four-layers/</link>
      <pubDate>Sun, 19 Jul 2026 09:00:00 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 00:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/super-individual-stack-four-layers/</guid>
      <description>用四层 AI 装备栈重新理解超级个体的长期优势：从最容易标准化的生产层，向判断、分发和声誉逐层上行，区分维持竞争资格的工具与能够沉淀复利的个人资产；结合开发者生产力研究的真实边界，给出判断当前瓶颈、停止过度配置工具、把每次交付转化为决策规则、读者关系和公开信誉的实践方法，并形成可持续的个人工作系统。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Super Individual</category>
      <category domain="tag">Solo Builder</category>
      <category domain="tag">Product Strategy</category>
      <category domain="tag">Productivity</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/super-individual-stack-four-layers.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/super-individual-stack-four-layers.jpeg" medium="image"><media:description>超级个体四层装备栈，从 AI 生产工具向判断、分发与声誉逐层上升</media:description></media:content>
    </item>
    <item>
      <title>你的 Mac 上有几套工具在打架？我设计了一个开发机体检 Skill</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/devbox-doctor-design/</link>
      <pubDate>Sat, 18 Jul 2026 13:30:00 +0800</pubDate>
      <atom:updated>Sat, 18 Jul 2026 13:30:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/devbox-doctor-design/</guid>
      <description>复盘 devbox-doctor 如何只读盘点 macOS 应用与常驻资源，用多源证据区分重复安装、配置冲突、卸载残留和不确定项，并限制模型只做提案。文章梳理 Spotlight 边界、Homebrew 语义、残留置信度、权限降级和误报评测，规定软件推荐必须附官方来源、版本与核验日期，适合想安全治理开发机的程序员。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Context Engineering</category>
      <category domain="tag">Automation</category>
      <category domain="tag">DevOps</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/devbox-doctor-design.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/devbox-doctor-design.jpeg" medium="image"><media:description>你的 Mac 上有几套工具在打架？我设计了一个开发机体检 Skill</media:description></media:content>
    </item>
    <item>
      <title>Agent Skill 与 SKILL.md 设计：我拆了一个敢删文件的插件</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/designing-valuable-agent-skills/</link>
      <pubDate>Sat, 18 Jul 2026 00:30:00 +0800</pubDate>
      <atom:updated>Sat, 18 Jul 2026 00:30:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/designing-valuable-agent-skills/</guid>
      <description>从 storage-analyzer 的 SKILL.md 与源码出发，拆解 Agent Skill 的需求筛选、代码—模型—代码架构、触发描述、结构化契约、风险分级和权限白名单。结合 macOS 真机测试，分析后台进程、输出缓冲与挂载重复计数，并给出可复用的设计、审查和验收清单，适合开发高风险 Skill 的工程师。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Context Engineering</category>
      <category domain="tag">Automation</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/designing-valuable-agent-skills.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/designing-valuable-agent-skills.jpeg" medium="image"><media:description>Agent Skill 与 SKILL.md 设计：我拆了一个敢删文件的插件</media:description></media:content>
    </item>
    <item>
      <title>搭建 AI 知识库：从零散笔记到可执行工作流</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/info-to-creation-arsenal/</link>
      <pubDate>Fri, 17 Jul 2026 21:30:00 +0800</pubDate>
      <atom:updated>Fri, 17 Jul 2026 21:30:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/info-to-creation-arsenal/</guid>
      <description>本文记录我如何把零散笔记建成一座可执行的 AI 知识库：用真实战绩约束方法卡，用目录和 Skill 装配创作上下文，用人工验收、失败日志与三日和七日复盘抵抗模型漂移。适合希望让个人知识管理真正进入内容工作流，又不愿把自动化误当成可靠性的创作者与工程团队。核心不是让 AI 代替判断，而是让每条方法可追溯、可比较、可撤回。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Context Engineering</category>
      <category domain="tag">Personal Growth</category>
      <category domain="tag">Automation</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/info-to-creation-arsenal.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/info-to-creation-arsenal.jpeg" medium="image"><media:description>连接证据卡、可执行工作流、内容平台与反馈回路的私有 AI 知识兵工厂</media:description></media:content>
    </item>
    <item>
      <title>一个人带一队 Agent，把 120 多篇文章的博客重新组装了一遍</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/ai-native-blog-rebuild/</link>
      <pubDate>Fri, 17 Jul 2026 17:30:00 +0800</pubDate>
      <atom:updated>Fri, 17 Jul 2026 17:30:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/ai-native-blog-rebuild/</guid>
      <description>我在 2026 年把维护四年、积累 120 多篇文章的 Hugo 博客改造成由人把关、Agent 执行的内容系统。本篇复盘母版流水线、Skills、自建 MCP、两段式封面、GitHub Actions SEO 日报与 GEO 实验，也交代权限边界、成本、失败案例和未完成事项，供维护知识库或内容站点的独立创作者参考。
</description>
      <category domain="category">Technology</category>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">MCP</category>
      <category domain="tag">Automation</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">GEO</category>
      <category domain="tag">Blog</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/ai-native-blog-rebuild.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/ai-native-blog-rebuild.jpeg" medium="image"><media:description>一个人带一队 Agent，把 120 多篇文章的博客重新组装了一遍</media:description></media:content>
    </item>
    <item>
      <title>2026 AI Agent 红海与蓝海：垂直工作流、支付协议与责任护城河</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/ai-agent-red-ocean-blue-ocean-2026/</link>
      <pubDate>Wed, 15 Jul 2026 18:00:00 +0800</pubDate>
      <atom:updated>Wed, 15 Jul 2026 18:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/ai-agent-red-ocean-blue-ocean-2026/</guid>
      <description>2026 年，AI Agent 的机会已从通用功能迁向垂直工作流、可信运行时、评测与支付协议。本文结合 Visa、Mastercard、AP2、Stripe、Munich Re 与 FIS 的一手进展，拆解专有数据、领域深度和端到端责任如何形成护城河，并给出判断红蓝海、审视产品壁垒与选择长期战场的可执行框架。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Product Strategy</category>
      <category domain="tag">Solo Builder</category>
      <category domain="tag">Super Individual</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/ai-agent-red-ocean-blue-ocean-2026.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/ai-agent-red-ocean-blue-ocean-2026.jpeg" medium="image"><media:description>2026 AI Agent 红海与蓝海：垂直工作流、支付协议与责任护城河</media:description></media:content>
    </item>
    <item>
      <title>如何建立对无人值守 AI Agent 的真实信任</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/trusting-unattended-ai-agent/</link>
      <pubDate>Wed, 15 Jul 2026 16:00:00 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 00:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/trusting-unattended-ai-agent/</guid>
      <description>本文给出无人值守 AI Agent 的工程化信任框架：以工具层最小权限护栏限制动作，用事故驱动的回归评测检验结果，以检查点、幂等恢复和分级人工审批控制复合风险，并说明如何用召回率、误报率、队列时延、成本与事故率持续校准系统。适合准备把 Agent 接入真实生产流程、又不愿把安全寄托在模型自觉上的工程团队与独立开发者。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Monitoring</category>
      <category domain="tag">Security</category>
      <category domain="tag">Testing</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/trusting-unattended-ai-agent.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/trusting-unattended-ai-agent.jpeg" medium="image"><media:description>夜间无人值守 Agent 工作流与晨间人工审批闸口</media:description></media:content>
    </item>
    <item>
      <title>超级个体的情报系统：从信源监测到可靠行动</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/super-individual-intelligence-system/</link>
      <pubDate>Wed, 15 Jul 2026 14:30:00 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 23:30:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/super-individual-intelligence-system/</guid>
      <description>面向超级个体的个人情报系统实战：把 RSS、RSSHub、网页监测、搜索 Agent 与 MCP 接成可运营流水线，保留信源证据，量化去重、摘要、延迟、成本和漏报，并用最小权限、人工审批、回滚与审计日志守住安全边界，让信息最终落到可解释的判断和行动，并在真实反馈中持续校正信源、阈值、摘要质量与自己的判断。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">MCP</category>
      <category domain="tag">Automation</category>
      <category domain="tag">Super Individual</category>
      <category domain="tag">Context Engineering</category>
      <category domain="tag">Monitoring</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/super-individual-intelligence-system.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/super-individual-intelligence-system.jpeg" medium="image"><media:description>超级个体的个人情报流水线，从信源监测连接到证据、判断与行动</media:description></media:content>
    </item>
    <item>
      <title>一个人养得起多大的 Agent 舰队：2026 年低价 API 与开放权重模型的成本账</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/open-model-cost-collapse-agent-fleet/</link>
      <pubDate>Wed, 15 Jul 2026 14:00:00 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 00:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/open-model-cost-collapse-agent-fleet/</guid>
      <description>本文以 2026 年 7 月 31 日官方价为基准，复算五千万 token 的 Agent 舰队成本，区分低价托管 API 与开放权重自托管两条路线，并把成功率、重试、工具、人审和严重错误纳入每个成功任务的总成本。适合需要多模型路由、控制上下文增长并评估个人 Agent 系统真实投入的开发者深入阅读。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Open Source</category>
      <category domain="tag">Super Individual</category>
      <category domain="tag">Solo Builder</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/open-model-cost-collapse-agent-fleet.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/open-model-cost-collapse-agent-fleet.jpeg" medium="image"><media:description>一场衡量 Agent 舰队真实成本的路由实验</media:description></media:content>
    </item>
    <item>
      <title>当 AI Agent 开始反过来提示你，真正改变了什么</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/proactive-agent-it-prompts-you/</link>
      <pubDate>Wed, 15 Jul 2026 12:00:00 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 00:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/proactive-agent-it-prompts-you/</guid>
      <description>主动式 AI Agent 正从等待指令走向在合适时机提出建议。本文以持久记忆、隔离运行时与事件触发三块底座，拆解过夜工作流、打扰阈值和反馈指标，并区分主动提示与自主决策，说明高风险动作为何仍需人工确认。适合设计 Agent 产品、自动化流程或工作系统的人阅读，结论是：真正的护城河不是更敢开口，而是知道何时保持沉默。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Product Strategy</category>
      <category domain="tag">Context Engineering</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/proactive-agent-it-prompts-you.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/proactive-agent-it-prompts-you.jpeg" medium="image"><media:description>当 AI Agent 开始反过来提示你，真正改变了什么</media:description></media:content>
    </item>
    <item>
      <title>让 AI 自动帮你追全网资讯，最后会卡在哪里</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/ai-auto-news-pipeline-limits/</link>
      <pubDate>Wed, 15 Jul 2026 10:00:00 +0800</pubDate>
      <atom:updated>Wed, 15 Jul 2026 10:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/ai-auto-news-pipeline-limits/</guid>
      <description>本文基于明确测试窗口内对 AI 资讯流水线的实测，拆解订阅聚合、变更监控与主动检索三类路线，校准 URL 归一、内容指纹、向量近重和长链可靠性的边界。文章给出带采样口径的去重漏斗、判断命中率记录法与人机协作清单，帮助持续跟踪技术变化的读者降低噪声，并说明缺少个人目标、上下文和反馈时，AI 为何不能可靠替代最终判断。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Automation</category>
      <category domain="tag">Super Individual</category>
      <category domain="tag">Monitoring</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/ai-auto-news-pipeline-limits.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/ai-auto-news-pipeline-limits.jpeg" medium="image"><media:description>让 AI 自动帮你追全网资讯，最后会卡在哪里</media:description></media:content>
    </item>
    <item>
      <title>从 Chatbot 到 Agent 到 Skill：AI 落地传统行业的真正分水岭</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/from-chatbot-to-agent-to-skill/</link>
      <pubDate>Tue, 14 Jul 2026 10:00:00 +0800</pubDate>
      <atom:updated>Tue, 14 Jul 2026 10:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/from-chatbot-to-agent-to-skill/</guid>
      <description>本文用 Chatbot→Agent→Skill 这套操作性框架，解释如何把一次性 AI 问答沉淀为可复用、可验证的业务流程；结合 Amazon Ads 官方边界与匿名跨境运营观察，拆解传统行业的机会、平台立场风险和人工审批责任，并给出落地的 Skill 模板与验证指标，适合把领域经验转成团队资产的从业者与独立开发者。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Context Engineering</category>
      <category domain="tag">Product Strategy</category>
      <category domain="tag">Automation</category>
      <enclosure url="https://cubxxw.com/images/blog/from-chatbot-to-agent-to-skill.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/blog/from-chatbot-to-agent-to-skill.svg" medium="image"><media:description>从 Chatbot 到 Agent 到 Skill——AI 落地的三级跃迁</media:description></media:content>
    </item>
    <item>
      <title>给 AI 任务，不只给方向：用完成态把协作落到结果</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/give-ai-tasks-not-directions/</link>
      <pubDate>Sat, 11 Jul 2026 16:30:00 +0800</pubDate>
      <atom:updated>Sat, 11 Jul 2026 16:30:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/give-ai-tasks-not-directions/</guid>
      <description>与 AI 协作时，方向适合探索，任务负责落地。本文用一张任务卡和完整案例，说明人如何守住价值取舍、责任与完成态，让 AI 在明确目标、输入、约束、输出、验收与停止条件下工作，用可检查的结果取代漫无边际的对话；也讨论何时可与 AI 共创方向，何时必须由人拍板，以及如何用多轮试验与人工复核减少偶然性，覆盖写作、研究与开发。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Context Engineering</category>
      <category domain="tag">Productivity</category>
      <category domain="tag">Learning</category>
      <enclosure url="https://cubxxw.com/images/blog/give-ai-tasks-not-directions.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/blog/give-ai-tasks-not-directions.svg" medium="image"><media:description>给 AI 一张有目标、约束与验收标准的任务卡</media:description></media:content>
    </item>
    <item>
      <title>AI 工作流质量门禁：把软件工程的纪律搬进个人系统</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/engineering-discipline-ai-workflow/</link>
      <pubDate>Sat, 11 Jul 2026 15:30:00 +0800</pubDate>
      <atom:updated>Sat, 11 Jul 2026 15:30:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/engineering-discipline-ai-workflow/</guid>
      <description>个人 AI 工作流缺的，并非更多提示词，关键在于一套能发现错误、保存证据并在高风险节点升级给人的质量门禁。本文从可观测性、指令层级、项目规则、持久记忆、评测与 SOP 出发，给出一份可直接复制的五项最小模板，帮助小团队降低模型自信出错、工具调用失真和流程过时的风险，把偶然可用的结果变成可审计、可复现、可持续改进的系统。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Testing</category>
      <category domain="tag">DevOps</category>
      <enclosure url="https://cubxxw.com/images/blog/engineering-discipline-ai-workflow.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/blog/engineering-discipline-ai-workflow.svg" medium="image"><media:description>AI 工作流质量门禁的四道工程纪律</media:description></media:content>
    </item>
    <item>
      <title>AI 第二大脑实操：Obsidian、Agent 与微信入口</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/ai-second-brain-build/</link>
      <pubDate>Sat, 11 Jul 2026 15:00:00 +0800</pubDate>
      <atom:updated>Sat, 11 Jul 2026 15:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/ai-second-brain-build/</guid>
      <description>用 Obsidian 管理本地 Markdown，以 Claude Code 或 Hermes 操作知识库，再把 QClaw 微信端作为远程入口。本指南给出 PARA 目录、权限边界、Git 备份、变更预览、分层验收与回滚步骤，并说明知识淘汰如何贯穿采集、加工和归档，帮助创作者先跑通工作流，再逐步扩展 AI 第二大脑。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Automation</category>
      <category domain="tag">Git</category>
      <category domain="tag">Security</category>
      <enclosure url="https://cubxxw.com/images/blog/ai-second-brain-build.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/blog/ai-second-brain-build.svg" medium="image"><media:description>Obsidian、Agent 与微信入口组成的三层 AI 第二大脑</media:description></media:content>
    </item>
    <item>
      <title>AI 内容创作工作流：把知识重组成读者愿意接收的作品</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/info-to-creation-creation/</link>
      <pubDate>Sat, 11 Jul 2026 14:40:00 +0800</pubDate>
      <atom:updated>Sat, 11 Jul 2026 14:40:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/info-to-creation-creation/</guid>
      <description>本文给出一套可复用的 AI 内容创作工作流：把知识卡依次经过受众任务、核心主张、证据、载体、人工验收、发布与反馈筛选，重组成读者愿意接收的作品。文章包含多载体案例、任务合同、责任矩阵、失败边界和发布指标，适合希望借 AI 提速，同时保留事实责任、作者声音与最终判断的创作者和工程团队，并让每次复盘回流为新知识。
</description>
      <category domain="category">Development</category>
      <category domain="tag">Product Strategy</category>
      <category domain="tag">Personal Growth</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Self-Discovery</category>
      <category domain="tag">Super Individual</category>
      <category domain="tag">Automation</category>
      <enclosure url="https://cubxxw.com/images/columns/info-to-creation/zh-05-creation.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/columns/info-to-creation/zh-05-creation.svg" medium="image"><media:description>知识卡经过 AI 辅助创作流水线抵达受众，再沿着可测量的反馈回路返回知识库</media:description></media:content>
    </item>
    <item>
      <title>AI 知识库工作流：把记录变成可验证、可复用的能力</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/info-to-creation-knowledge/</link>
      <pubDate>Sat, 11 Jul 2026 14:30:00 +0800</pubDate>
      <atom:updated>Sat, 11 Jul 2026 14:30:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/info-to-creation-knowledge/</guid>
      <description>本文给出一套 AI 知识库工作流：让记录经过证据门、状态迁移、显式检索、人工复查与可逆退役，变成可验证、可复用的能力沉淀。文章包含完整知识卡、入库协议、PARA 边界、替代索引、检索失败日志、墓碑机制和健康指标，适合希望让知识真正参与 AI 上下文，又不愿把目录整齐误当成模型理解的创作者与工程团队。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Context Engineering</category>
      <category domain="tag">RAG</category>
      <category domain="tag">Automation</category>
      <enclosure url="https://cubxxw.com/images/columns/info-to-creation/zh-04-knowledge.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/columns/info-to-creation/zh-04-knowledge.svg" medium="image"><media:description>草稿记录经过证据门成为可检索知识卡，再沿复查与退役回路返回更新</media:description></media:content>
    </item>
    <item>
      <title>AI 记录工作流：把易逝输入变成可验证的半成品</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/info-to-creation-record/</link>
      <pubDate>Sat, 11 Jul 2026 14:20:00 +0800</pubDate>
      <atom:updated>Sat, 11 Jul 2026 14:20:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/info-to-creation-record/</guid>
      <description>本文给出一套 AI 记录工作流：把文字、语音、截图、代码变更、实验输出和决策日志变成可追溯的半成品，再经过澄清、人工复查与证据门晋级知识。文章包含五种状态、最小捕获卡、敏感数据边界、八项复盘、脱敏案例和质量指标，适合希望降低记录摩擦，又不愿丢掉来源、上下文与责任的创作者和工程团队。重点是可重建，不是记得多。
</description>
      <category domain="category">Development</category>
      <category domain="tag">Personal Growth</category>
      <category domain="tag">Self-Discovery</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Automation</category>
      <category domain="tag">Product Strategy</category>
      <enclosure url="https://cubxxw.com/images/columns/info-to-creation/zh-03-record.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/columns/info-to-creation/zh-03-record.svg" medium="image"><media:description>文字、语音、截图、代码变更与决策依次经过捕获、澄清和复查，晋级为可验证知识卡</media:description></media:content>
    </item>
    <item>
      <title>AI 信息筛选工作流：抓住信号，不建噪音仓库</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/info-to-creation-information/</link>
      <pubDate>Sat, 11 Jul 2026 14:10:00 +0800</pubDate>
      <atom:updated>Sat, 11 Jul 2026 14:10:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/info-to-creation-information/</guid>
      <description>本文给出一套 AI 信息筛选工作流：用任务相关性、来源质量、时效、独特性、敏感度、版权、成本和可复核性守住入口，并区分可交给 AI、先脱敏、仅人工处理、限时浏览与拒绝。包含来源护照、提示注入防线、脱敏案例、摘要验收和抽检指标，适合不想把收藏夹变成噪音仓库的创作者与工程团队。重点是守好入口，不是继续囤积。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Personal Growth</category>
      <category domain="tag">Self-Discovery</category>
      <category domain="tag">Product Strategy</category>
      <category domain="tag">Automation</category>
      <enclosure url="https://cubxxw.com/images/columns/info-to-creation/zh-02-information.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/columns/info-to-creation/zh-02-information.svg" medium="image"><media:description>来源依次经过相关性、隐私与核验闸门，再由 AI 辅助处理并进入记录层</media:description></media:content>
    </item>
    <item>
      <title>AI 知识工作流：从信息、记录、知识到创作</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/info-to-creation-the-framework/</link>
      <pubDate>Sat, 11 Jul 2026 14:00:00 +0800</pubDate>
      <atom:updated>Sat, 11 Jul 2026 14:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/info-to-creation-the-framework/</guid>
      <description>本文给出一套完整的 AI 知识工作流：让信息依次经过入口海关、可追溯记录、证据化知识和面向受众的创作，并通过反馈回流持续修正。文章统一四层输入、状态、AI 角色、人工责任、验收标准和健康指标，适合希望借 AI 提速，又不愿丢掉来源、范围、判断与发布责任的创作者和工程团队。重点不是存得多，而是交接可查。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Personal Growth</category>
      <category domain="tag">Context Engineering</category>
      <category domain="tag">Product Strategy</category>
      <category domain="tag">Self-Discovery</category>
      <category domain="tag">Automation</category>
      <enclosure url="https://cubxxw.com/images/columns/info-to-creation/zh-01-framework.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/columns/info-to-creation/zh-01-framework.svg" medium="image"><media:description>信息经过入口闸门成为记录和知识，再面向受众创作并沿可测量反馈回路返回</media:description></media:content>
    </item>
    <item>
      <title>2026 GEO 度量指南：可复现的 AI 引用与转介监测方法</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/geo-measurement-and-tools/</link>
      <pubDate>Sat, 11 Jul 2026 12:00:00 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 00:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/geo-measurement-and-tools/</guid>
      <description>建立可复现的 GEO 度量体系：区分生成式搜索曝光、提及与有效引用、AI 转介访问及最终转化，说明 2026 年 Google Search Console 生成式 AI 报告、GA4 归因和提示词审计的能力边界，并给出适合个人博客落地的记录协议、分母定义与工具选择方法，避免把代理指标误当成真实效果。
</description>
      <category domain="category">Development</category>
      <category domain="tag">GEO</category>
      <category domain="tag">SEO</category>
      <category domain="tag">AI Search</category>
      <category domain="tag">Content Strategy</category>
      <category domain="tag">Analytics</category>
      <category domain="tag">LLM</category>
      <enclosure url="https://cubxxw.com/images/columns/geo/zh-06-measurement.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/columns/geo/zh-06-measurement.svg" medium="image"><media:description>GEO 度量框架封面，连接平台曝光、有效引用、转介访问与最终转化</media:description></media:content>
    </item>
    <item>
      <title>GEO 博客改造复盘：用真实数据重跑五层模型</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/geo-blog-rebuild-case-study/</link>
      <pubDate>Sat, 11 Jul 2026 11:30:00 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 00:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/geo-blog-rebuild-case-study/</guid>
      <description>这是一篇基于真实数据的 GEO 博客改造复盘：我用 Search Console、Lighthouse 与本地审计拆解 87.8 万曝光背后的噪声和有效需求，重新核对 nsddd.top 到 cubxxw.com 的域名迁移，并按风险、证据与可验证性排出 Hugo 博客的重建顺序，同时说明这些数字不能证明什么。
</description>
      <category domain="category">Development</category>
      <category domain="tag">GEO</category>
      <category domain="tag">SEO</category>
      <category domain="tag">AI Search</category>
      <category domain="tag">Content Strategy</category>
      <category domain="tag">Blog</category>
      <enclosure url="https://cubxxw.com/images/columns/geo/zh-05-case-study.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/columns/geo/zh-05-case-study.svg" medium="image"><media:description>GEO 博客改造复盘封面，展示真实数据仪表盘与增长曲线</media:description></media:content>
    </item>
    <item>
      <title>GEO 站外信任：身份、证据与合乎伦理的社区分发</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/geo-trust-and-endorsement/</link>
      <pubDate>Sat, 11 Jul 2026 11:00:00 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 12:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/geo-trust-and-endorsement/</guid>
      <description>本文以 Google 官方文档和 2026 年跨语言研究为边界，说明个人博客如何通过作者身份、第三方证据与合乎伦理的社区分发建立站外信任，并用固定问题集、重复运行、引用支持度和读者结果检验变化。适合希望提升 GEO 可见度，又不愿把 Reddit、Wikipedia 或 sameAs 当成排名捷径的作者与工程团队。
</description>
      <category domain="category">Development</category>
      <category domain="tag">GEO</category>
      <category domain="tag">AI Search</category>
      <category domain="tag">Content Strategy</category>
      <category domain="tag">SEO</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <enclosure url="https://cubxxw.com/images/columns/geo/zh-04-trust.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/columns/geo/zh-04-trust.svg" medium="image"><media:description>连接作者、原创成果、独立证据与读者社区的可验证信任网络</media:description></media:content>
    </item>
    <item>
      <title>GEO 结构化内容实战：先服务读者，再验证 AI 可见度</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/geo-structured-content-tactics/</link>
      <pubDate>Sat, 11 Jul 2026 10:30:00 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 18:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/geo-structured-content-tactics/</guid>
      <description>一份面向技术写作者的 GEO 结构化内容指南：从真实读者问题出发，写清结论、证据、适用范围和下一步；正确理解问题式标题、FAQ/HowTo Schema、llms.txt、摘要与内链的边界；最后用固定问题、平台、日期、地区和实际引用记录，验证 AI 可见度是否发生变化，不把排版技巧包装成确定的排名或引用承诺。
</description>
      <category domain="category">Development</category>
      <category domain="tag">GEO</category>
      <category domain="tag">Content Strategy</category>
      <category domain="tag">SEO</category>
      <category domain="tag">AI Search</category>
      <category domain="tag">Hugo</category>
      <enclosure url="https://cubxxw.com/images/columns/geo/zh-03-structured.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/columns/geo/zh-03-structured.svg" medium="image"><media:description>GEO 结构化内容实战封面，展示清晰且可复用的内容块</media:description></media:content>
    </item>
    <item>
      <title>AI 搜索如何检索与引用来源：一套可验证的 GEO 模型</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/geo-how-ai-retrieves-and-cites/</link>
      <pubDate>Sat, 11 Jul 2026 10:00:00 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 12:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/geo-how-ai-retrieves-and-cites/</guid>
      <description>本文以 2026 年公开资料和可复现实验为边界，拆解 Google、Perplexity 与 ChatGPT 的检索和引用机制，区分可访问、被选择、被吸收、被引用与有用五个结果，并给出固定问题集、环境记录、重复运行和单变量修改的方法。适合希望提升 AI 搜索可见度、又不愿被 GEO 魔法数字误导的内容作者与工程团队。
</description>
      <category domain="category">Development</category>
      <category domain="tag">GEO</category>
      <category domain="tag">RAG</category>
      <category domain="tag">AI Search</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Content Strategy</category>
      <enclosure url="https://cubxxw.com/images/columns/geo/zh-02-retrieval.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/columns/geo/zh-02-retrieval.svg" medium="image"><media:description>AI 搜索中可访问、来源选择、答案吸收、可见引用与用户价值的可测链路</media:description></media:content>
    </item>
    <item>
      <title>2026 GEO 实证指南：证据、边界与可执行工作流</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/geo-generative-engine-optimization-guide/</link>
      <pubDate>Fri, 10 Jul 2026 22:00:00 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 18:43:21 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/geo-generative-engine-optimization-guide/</guid>
      <description>一份以证据为边界的 2026 GEO 实践指南：根据 Google、OpenAI、Perplexity 官方文档和 Pew、Bain、KDD 原始研究，说明生成式搜索的技术基础、crawler 权限、内容证据与测量方法，区分可见度、引用、访问和转化，避免把 llms.txt、Schema 或二手统计包装成排名捷径。
</description>
      <category domain="category">Development</category>
      <category domain="tag">GEO</category>
      <category domain="tag">SEO</category>
      <category domain="tag">AI Search</category>
      <category domain="tag">Content Strategy</category>
      <category domain="tag">LLM</category>
      <enclosure url="https://cubxxw.com/images/columns/geo/zh-01-guide.svg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/columns/geo/zh-01-guide.svg" medium="image"><media:description>把访问、独特价值、证据、可摘录性和度量分开的五层 GEO 工作流</media:description></media:content>
    </item>
    <item>
      <title>open-lovable 源码拆解：Agentic Search、文本协议与双沙箱</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/dissecting-open-lovable/</link>
      <pubDate>Mon, 29 Jun 2026 09:30:00 +0800</pubDate>
      <atom:updated>Mon, 29 Jun 2026 09:30:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/dissecting-open-lovable/</guid>
      <description>基于固定提交审计 Firecrawl 的 open-lovable，拆解网址抓取、流式代码生成、Agentic Search、Morph 增量编辑与 E2B、Vercel 双沙箱实现。文章区分仓库当时使用的能力与平台当前提供的功能，纠正命中率、耗时和隔离层误读，适合设计 AI 应用生成器、代码检索与执行边界的开发者。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Architecture</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Development</category>
      <category domain="tag">Open Source</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/dissecting-open-lovable.png" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/dissecting-open-lovable.png" medium="image"><media:description>open-lovable 从网页抓取到代码生成与双沙箱预览的源码架构</media:description></media:content>
    </item>
    <item>
      <title>Relay Agent 架构审计：从设计承诺到本地实现</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/relay-agent-architecture-design/</link>
      <pubDate>Wed, 24 Jun 2026 10:00:00 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 18:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/relay-agent-architecture-design/</guid>
      <description>基于 Relay 私有本地实现的源码审计，核对协调器、五个领域 Agent、Harness、路由、事件系统与浏览器投递链路，区分已实现、已测试、设计中和示意内容，并重点解释 LangGraph 恢复语义、幂等边界、审计持久化、反虚构守卫的能力边界，以及浏览器自动化面对服务条款、提示注入和敏感字段时保留的人类审批机制。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Python</category>
      <category domain="tag">Security</category>
      <category domain="tag">Development</category>
      <enclosure url="https://cubxxw.com/images/blog/relay-agent-architecture.webp" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/blog/relay-agent-architecture.webp" medium="image"><media:description>Relay 协调器、五个领域 Agent、浏览器投递、防护机制和审计边界组成的架构图</media:description></media:content>
    </item>
    <item>
      <title>Context 不是 Prompt：上下文工程如何成为 AI Agent 的新地基</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/context-engineering-the-new-foundation/</link>
      <pubDate>Mon, 22 Jun 2026 03:30:00 +0800</pubDate>
      <atom:updated>Mon, 22 Jun 2026 03:30:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/context-engineering-the-new-foundation/</guid>
      <description>本文从 Anthropic 与 Sourcegraph 的代表性框架出发，解释上下文工程与提示词工程的边界，并核对上下文腐烂、检索重排、Claude Code 自动压缩、服务端 compaction 与记忆系统。面向 AI Agent 工程师；核心结论是上下文并非越多越好，关键是让高信号信息在正确时刻进入窗口。
</description>
      <category domain="category">Development</category>
      <category domain="tag">Context Engineering</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Agent</category>
      <category domain="tag">MCP</category>
      <enclosure url="https://cubxxw.com/images/blog/context-engineering-desk.webp" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/blog/context-engineering-desk.webp" medium="image"><media:description>书桌上的笔记本电脑显示「Context 不是 Prompt」文章与核心要点面板，旁边摊开的笔记写着 Context Engineering 的四支柱与 Worldline</media:description></media:content>
    </item>
    <item>
      <title>Agent Engineering 全景地图：98.4% 只是 Harness 的叙事锚点</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/agent-engineering-the-98-percent-harness/</link>
      <pubDate>Wed, 17 Jun 2026 09:30:00 +0800</pubDate>
      <atom:updated>Wed, 17 Jun 2026 09:30:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/agent-engineering-the-98-percent-harness/</guid>
      <description>从并非论文硬测量的 98.4% 说起，拆解 Agent Harness 的编排、上下文、记忆、工具、可靠性、评估、成本与治理八根支柱。文章以截至 2026 年 7 月的一手资料校准关键事实，给出每根支柱的最小实现、失效边界和选型判断，帮助工程师把会调用模型的演示，推进为可恢复、可评估、可治理的生产系统。
</description>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Context Engineering</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">MCP</category>
      <enclosure url="https://cubxxw.com/images/blog/agent-engineering-harness.webp" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/blog/agent-engineering-harness.webp" medium="image"><media:description>一张技术示意图，中心是一个小小的 agent loop，外面一圈一圈包裹着编排、上下文、记忆、工具、可靠性、评估、成本、治理八根支柱</media:description></media:content>
    </item>
    <item>
      <title>Agent 的自我：从洛克到 OpenClaw</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/agent-identity-from-locke-to-openclaw/</link>
      <pubDate>Sun, 05 Apr 2026 20:00:00 +0800</pubDate>
      <atom:updated>Sun, 05 Apr 2026 20:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/agent-identity-from-locke-to-openclaw/</guid>
      <description>本文从洛克《人类理解论》第二卷第二十七章出发，讨论 AI Agent 如何跨会话保持行为连续性。核验 OpenClaw、Mem0、Karpathy LLM Wiki、SoulSpec 与 EvoMap 一手资料，给出身份文件、运行记忆、权限边界、经验谱系和评测设计清单，面向构建长期智能体、关注审计与信任边界的工程师。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Context Engineering</category>
      <category domain="tag">Harness Engineering</category>
      <category domain="tag">Philosophy</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2026/agent-identity-from-locke-to-openclaw.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2026/agent-identity-from-locke-to-openclaw.jpeg" medium="image"><media:description>Agent 身份连续性的文件、记忆、运行环境与评测结构</media:description></media:content>
    </item>
    <item>
      <title>Mem0 OSS v3 实践：记忆架构、混合检索与生产取舍</title>
      <link>https://cubxxw.com/zh/projects/mem0/</link>
      <pubDate>Fri, 09 May 2025 21:33:46 +0800</pubDate>
      <atom:updated>Fri, 09 May 2025 21:33:46 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/projects/mem0/</guid>
      <description>基于 Mem0 OSS v3，拆解 ADD-only 写入、语义、BM25 与实体信号融合检索、隐私治理和评测，并厘清开源版与 Platform Graph Memory 的边界。帮助需要跨会话个性化的团队判断何时采用、如何迁移，何时改用数据库或 LangGraph checkpointer，避免把概率记忆当成事实。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Open Source</category>
      <category domain="tag">Project Learning</category>
      <category domain="tag">RAG</category>
      <category domain="tag">LLM</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2025/mem0.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2025/mem0.jpeg" medium="image"><media:description>低饱和档案卡片以线索相连，象征 Mem0 OSS v3 的选择性长期记忆</media:description></media:content>
    </item>
    <item>
      <title>UFO² 桌面 AgentOS：从 Windows 自动化到 UFO³ Galaxy</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/UFO/</link>
      <pubDate>Fri, 09 May 2025 21:30:15 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 01:41:06 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/UFO/</guid>
      <description>从 UFO v1 的界面操作，到 UFO² 的 Windows 桌面 AgentOS，再到 UFO³ Galaxy 的跨设备编排，本文以 UFO² 为主线拆解 HostAgent、AppAgent、GUI 与 API 混合执行、知识检索、PiP 隔离桌面和 MCP，并给出 2026 年仍可使用的安装配置方法与工程判断。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Automation</category>
      <category domain="tag">Development</category>
      <category domain="tag">MCP</category>
      <category domain="tag">Open Source</category>
      <category domain="tag">Project Learning</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2025/UFO.png" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2025/UFO.png" medium="image"><media:description>UFO² 桌面 AgentOS 的多智能体架构</media:description></media:content>
    </item>
    <item>
      <title>LLM 推荐系统工程指南：语义召回、排序、强化学习与评估</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/ai-recommend/</link>
      <pubDate>Wed, 23 Apr 2025 10:39:05 +0800</pubDate>
      <atom:updated>Wed, 23 Apr 2025 10:39:05 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/ai-recommend/</guid>
      <description>本文梳理大语言模型进入推荐系统的工程路径：融合协同过滤与语义表示，设计召回、排序、重排和反馈闭环，并辨析 SeLLa-Rec、Semantic ID、Rec-R1、UQABench 的证据边界。文章给出数据、延迟、离线评估和 A/B 测试清单，帮助团队判断何时使用 LLM、何时保留经典模型，避免被实验指标误导。
</description>
      <category domain="category">Technology</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">RAG</category>
      <category domain="tag">Monitoring</category>
      <category domain="tag">Project Learning</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2025/ai-recommend.png" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2025/ai-recommend.png" medium="image"><media:description>一组空白卡片经过木质筛选阶梯，最终陈列在用户座椅前</media:description></media:content>
    </item>
    <item>
      <title>Gemini Notebook（原 NotebookLM）：从来源到决策的研究工作流</title>
      <link>https://cubxxw.com/zh/projects/notebooklm/</link>
      <pubDate>Mon, 21 Apr 2025 22:59:57 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 12:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/projects/notebooklm/</guid>
      <description>本文以 2026 年 7 月 Google 官方资料为准，说明 NotebookLM 更名为 Gemini Notebook 后的来源边界、引用核验、研究模式、Studio 输出、套餐、移动端限制与隐私差异，并给出一套从问题、证据到决策的可复现流程。适合需要可靠溯源、避免把流畅答案误当结论的研究者、工程师与知识工作者。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">RAG</category>
      <category domain="tag">Productivity</category>
      <category domain="tag">Learning</category>
      <category domain="tag">Context Engineering</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2025/notebooklm.png" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2025/notebooklm.png" medium="image"><media:description>研究笔记中的来源页面、引用连线与决策卡片</media:description></media:content>
    </item>
    <item>
      <title>AI 与大模型应用的测试驱动开发：从确定性测试到风险化评测</title>
      <link>https://cubxxw.com/zh/projects/tdd/</link>
      <pubDate>Mon, 21 Apr 2025 15:52:34 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 10:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/projects/tdd/</guid>
      <description>面向 AI 与大模型应用的测试驱动开发指南：解释红—绿—重构在概率系统中的适用边界，搭建单元测试、契约集成、离线评测与线上监控四层防线，并给出 Python、React、Go 示例，以及评测器校准、风险分级 CI 和 Cursor Agent 审批策略，帮助团队在不掩盖不确定性的前提下建立可验证、可回滚的契约。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Testing</category>
      <category domain="tag">Python</category>
      <category domain="tag">Go</category>
      <category domain="tag">Automation</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2025/tdd.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2025/tdd.jpeg" medium="image"><media:description>红、绿、蓝三色测试闭环环绕一个受约束的 AI 系统</media:description></media:content>
    </item>
    <item>
      <title>MarkItDown 文档转 Markdown 实战：0.1.6 机制基线与 0.1.7 状态</title>
      <link>https://cubxxw.com/zh/projects/markitdown/</link>
      <pubDate>Mon, 21 Apr 2025 15:41:21 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 16:30:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/projects/markitdown/</guid>
      <description>这是一份面向开发者的 MarkItDown 文档摄取实战指南，以可复核的 0.1.6 机制为基线并说明 0.1.7 发布状态，讲清内置转换、OCR 插件、视觉模型、Azure 文档智能与内容理解的边界，提供选型矩阵、质量门和可复现实验，帮助你在进入 RAG 索引前发现阅读顺序、表格、图片文字与成本风险。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">Open Source</category>
      <category domain="tag">Python</category>
      <category domain="tag">LLM</category>
      <category domain="tag">RAG</category>
      <category domain="tag">Project Learning</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2025/markItdown.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2025/markItdown.jpeg" medium="image"><media:description>文档经过分层转换与质量检查后成为结构化 Markdown</media:description></media:content>
    </item>
    <item>
      <title>2026 LangGraph 架构指南：StateGraph、持久化与故障恢复</title>
      <link>https://cubxxw.com/zh/projects/langgraph/</link>
      <pubDate>Sat, 19 Apr 2025 15:19:20 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 18:11:30 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/projects/langgraph/</guid>
      <description>面向 2026 年生产实践的 LangGraph 架构指南：解释 StateGraph 的状态、节点与边，演示检查点、thread_id、interrupt 与 Command 恢复流程，并用退款审批故障路径说明幂等副作用、重试边界和持久化责任，帮助团队判断何时从 create_agent 下沉到 LangGraph。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Python</category>
      <category domain="tag">Open Source</category>
      <category domain="tag">Project Learning</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2025/langgraph.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2025/langgraph.jpeg" medium="image"><media:description>带检查点、人工审批与恢复路径的 LangGraph 状态图</media:description></media:content>
    </item>
    <item>
      <title>LangChain 1.x 生产实践：模型、Agent 与 LangGraph 如何选</title>
      <link>https://cubxxw.com/zh/projects/langchain/</link>
      <pubDate>Wed, 16 Apr 2025 17:36:46 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 10:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/projects/langchain/</guid>
      <description>这是一份面向生产环境的 LangChain 1.x 工程指南：从直接调用模型、create_agent 到 LangGraph，给出清晰选型边界、可运行的 RAG 与智能体示例、人机审批设计，以及评估、权限、状态和可观测性检查清单，帮助团队避开旧 API、过度抽象与不可控的自动化，引入框架时为长期维护保留清晰路径。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">Open Source</category>
      <category domain="tag">LangChain</category>
      <category domain="tag">Agent</category>
      <category domain="tag">RAG</category>
      <category domain="tag">Python</category>
      <category domain="tag">LLM</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2025/langchain.png" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2025/langchain.png" medium="image"><media:description>模型、智能体与图工作流沿三条路径汇合的克制编辑插画</media:description></media:content>
    </item>
    <item>
      <title>AI Gateway 选型指南：LiteLLM、Kong、APISIX、Cloudflare 与 Portkey 怎么选</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/ai-gateway/</link>
      <pubDate>Wed, 16 Apr 2025 17:36:12 +0800</pubDate>
      <atom:updated>Wed, 16 Apr 2025 17:36:12 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/ai-gateway/</guid>
      <description>本文从架构、路由与容错、缓存、预算治理、安全、可观测性六个维度，对比 LiteLLM、Kong、Apache APISIX、Cloudflare AI Gateway 与 Portkey，厘清开源版、托管版和企业版的能力边界，并给出适合创业团队、平台团队与强合规场景的选择方法、上线前验证清单及可迁移的落地建议。
</description>
      <category domain="category">Technology</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Cloud Native</category>
      <category domain="tag">Security</category>
      <category domain="tag">Monitoring</category>
      <category domain="tag">Open Source</category>
      <category domain="tag">DevOps</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2025/ai-gateway.png" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2025/ai-gateway.png" medium="image"><media:description>多条模型调用管道汇入 AI Gateway，并由量杯与天平衡量成本和治理</media:description></media:content>
    </item>
    <item>
      <title>独立开发者技术栈指南：从问题验证到可维护 MVP</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/independent-developer/</link>
      <pubDate>Tue, 15 Apr 2025 20:53:12 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 12:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/independent-developer/</guid>
      <description>这是一份写给独立开发者的常青技术栈与MVP路线图：从访谈和手工交付验证真实问题，收敛前后端、数据库与API选择，再补齐Git、Docker、持续集成、测试、部署、支付、分析和维护机制。文章不列容易过期的工具榜单，而用阶段决策表、退出成本和真实取舍，帮助一人团队少造基础设施、尽早收费，让产品保持可迁移、可观测、可演进。
</description>
      <category domain="category">Development</category>
      <category domain="tag">Solo Builder</category>
      <category domain="tag">Product Strategy</category>
      <category domain="tag">Development</category>
      <category domain="tag">Git</category>
      <category domain="tag">Docker</category>
      <category domain="tag">Testing</category>
      <category domain="tag">Deployment</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2025/independent-developer.png" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2025/independent-developer.png" medium="image"><media:description>独立开发者从问题验证走向可维护MVP的技术路线图</media:description></media:content>
    </item>
    <item>
      <title>GPT Researcher 源码审计：深度研究代理如何检索、写作与自托管</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/gpt-researcher/</link>
      <pubDate>Mon, 14 Apr 2025 16:17:27 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 10:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/gpt-researcher/</guid>
      <description>本文以 GPT Researcher v3.5.0（b364917）为源码基线，核对 Python 3.11、调用链、Docker、MCP 五项工具、LangSmith 与 DeepResearchGym，并厘清 Apache-2.0 代码、自托管资源和 API 成本，帮助工程师判断它是否适合生产研究流程。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">Project Learning</category>
      <category domain="tag">Open Source</category>
      <category domain="tag">Python</category>
      <category domain="tag">Docker</category>
      <category domain="tag">MCP</category>
      <category domain="tag">RAG</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2025/gpt-researcher.png" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2025/gpt-researcher.png" medium="image"><media:description>GPT Researcher 深度研究代理的检索、证据与报告生成流程</media:description></media:content>
    </item>
    <item>
      <title>Jina 2026：搜索底座模型、API 与 Jina Serve 实战指南</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/jina/</link>
      <pubDate>Sat, 12 Apr 2025 13:01:59 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 12:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/jina/</guid>
      <description>Elastic 收购 Jina AI 后，Jina 已从云原生神经搜索框架转向搜索底座模型品牌。本文按 2026 年现状厘清托管 API 与开源 Jina Serve 的边界，比较 v4、v5 text/omni、reranker v3.5 的上下文、维度和许可证，并给出 API、MCP、Flow 示例与选型矩阵。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">RAG</category>
      <category domain="tag">LLM</category>
      <category domain="tag">MCP</category>
      <category domain="tag">Python</category>
      <category domain="tag">Open Source</category>
      <category domain="tag">Project Learning</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2025/jina.png" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2025/jina.png" medium="image"><media:description>Jina 搜索底座模型、托管 API 与 Jina Serve 的两层架构</media:description></media:content>
    </item>
    <item>
      <title>LangChain 1.x 实战指南：从模型调用到可观测 Agent</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/harnessing-language-model-applications-with-langchain-a-developer-is-guide/</link>
      <pubDate>Wed, 22 May 2024 21:37:34 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 10:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/harnessing-language-model-applications-with-langchain-a-developer-is-guide/</guid>
      <description>面向 Python 开发者的 LangChain 1.x 实战指南，覆盖模型初始化、工具调用、结构化输出、中间件，以及 LangGraph 记忆、RAG、SQL Agent、Ollama 本地模型和 LangSmith 追踪。文章提供 API 迁移表、可运行代码与排错清单，帮助你避开过时示例，构建可维护的智能体应用。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Python</category>
      <category domain="tag">RAG</category>
      <category domain="tag">Development</category>
      <category domain="tag">Automation</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2024/harnessing-language-model-applications-with-langchain-a-developer-is-guide.png" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2024/harnessing-language-model-applications-with-langchain-a-developer-is-guide.png" medium="image"><media:description>LangChain 1.x 模型、Agent、记忆与可观测性工程架构</media:description></media:content>
    </item>
    <item>
      <title>从语言模型到 RAG：理解大模型的能力、边界与工程方法</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/exploring-large-language-models-llms-pioneering-ai-understanding-generation-human-language/</link>
      <pubDate>Wed, 15 May 2024 20:12:29 +0800</pubDate>
      <atom:updated>Wed, 15 May 2024 20:12:29 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/exploring-large-language-models-llms-pioneering-ai-understanding-generation-human-language/</guid>
      <description>从预测下一个词出发，理解 Transformer、规模化与所谓涌现能力，再走进检索增强生成的完整工程链路。本文不追逐模型榜单，而是解释大模型为何有效、何时会失败，如何设计检索、引用、评估与安全边界，帮助开发者把概率性的语言能力变成可验证、可维护、能持续演进的知识系统。 也给出一条从原型走向生产环境的务实路径。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">RAG</category>
      <category domain="tag">Context Engineering</category>
      <category domain="tag">Development</category>
      <category domain="tag">Testing</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2024/exploring-large-language-models-llms-pioneering-ai-understanding-generation-human-language.jpeg" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2024/exploring-large-language-models-llms-pioneering-ai-understanding-generation-human-language.jpeg" medium="image"><media:description>从语言模型到 RAG：理解大模型的能力、边界与工程方法</media:description></media:content>
    </item>
    <item>
      <title>Sora 停服之后：SoraEase 提示词档案与开发者迁移指南</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/sora-ease-guide-mastering-sora-ai-for-developers/</link>
      <pubDate>Thu, 14 Mar 2024 08:44:13 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 16:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/sora-ease-guide-mastering-sora-ai-for-developers/</guid>
      <description>这是一份写于 2024 年、在 Sora 停服后重新校订的 SoraEase 提示词档案。文章核对 Sora 从研究预览、正式发布到 API 退场的时间线，保留五个有方法价值的提示词，并给出场景契约、三层评测、适配器设计与退出清单，帮助仍在使用视频生成接口的开发者迁移，同时避免把历史经验误写成仍然有效的产品承诺。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">Development</category>
      <category domain="tag">Open Source</category>
      <category domain="tag">Automation</category>
      <category domain="tag">Testing</category>
      <category domain="tag">Content Strategy</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2024/sora-ease-guide-mastering-sora-ai-for-developers.png" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2024/sora-ease-guide-mastering-sora-ai-for-developers.png" medium="image"><media:description>废弃的胶片越过桥梁，连接到模块化视频工作流</media:description></media:content>
    </item>
    <item>
      <title>Sora 完整复盘：从视频生成技术预览、Sora 2 到产品与 API 停止</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/exploring-sora-technology-for-enthusiasts-and-developers/</link>
      <pubDate>Sat, 24 Feb 2024 13:30:15 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 00:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/exploring-sora-technology-for-enthusiasts-and-developers/</guid>
      <description>本文复盘Sora从2024年研究预览、Sora Turbo产品化，到2025年Sora 2加入同步音频，再到2026年产品停止、API进入关闭倒计时的路径；拆解扩散Transformer与时空patch，澄清一分钟研究样例与1080p、20秒产品规格的区别，并从内容溯源、肖像同意、版权风险与系统迁移中提炼开发者教训。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">Development</category>
      <category domain="tag">Product Strategy</category>
      <category domain="tag">Security</category>
      <category domain="tag">Content Strategy</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2024/exploring-sora-technology-for-enthusiasts-and-developers.png" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2024/exploring-sora-technology-for-enthusiasts-and-developers.png" medium="image"><media:description>Sora 从技术预览、产品化到停止服务的时间线</media:description></media:content>
    </item>
    <item>
      <title>向量数据库的学习</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/vector-database-learning/</link>
      <pubDate>Sat, 20 Jan 2024 12:57:15 +0800</pubDate>
      <atom:updated>Sat, 20 Jan 2024 12:57:15 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/vector-database-learning/</guid>
      <description>本文介绍了向量数据库的基础知识，使用场景，以及它在现代大数据和人工智能领域中的应用。</description>
      <category domain="tag">Blog</category>
      <category domain="tag">AI</category>
      <enclosure url="https://cubxxw.com/images/blog/vector-database.png" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/blog/vector-database.png" medium="image"></media:content>
    </item>
    <item>
      <title>2024 AI 现场笔记：从大模型涌现到生产级 RAG，2026 再校准</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/emerging-challenges-and-trends-in-2024/</link>
      <pubDate>Sun, 14 Jan 2024 22:52:24 +0800</pubDate>
      <atom:updated>Thu, 30 Jul 2026 22:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/emerging-challenges-and-trends-in-2024/</guid>
      <description>这是一份从2024年初大模型分享会出发、在2026年重新校准的工程笔记。文章复盘涌现争议、Prefix LM、LoRA、QLoRA及Agent项目，逐项标明哪些判断仍成立、哪些已经变化，并给出面向生产环境的RAG评估、引用溯源、权限隔离、重排、拒答与可观测性实践，帮助团队把模型演示变成可验证、可审计的知识系统。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">LLM</category>
      <category domain="tag">RAG</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Monitoring</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2024/emerging-challenges-and-trends-in-2024.png" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2024/emerging-challenges-and-trends-in-2024.png" medium="image"><media:description>从2024大模型现场笔记走向2026生产级RAG系统</media:description></media:content>
    </item>
    <item>
      <title>AutoGPT 2026：从 Classic 实验到 Platform 的迁移指南</title>
      <link>https://cubxxw.com/zh/ai-agent/posts/use-auto-gpt/</link>
      <pubDate>Sun, 16 Jul 2023 16:28:31 +0800</pubDate>
      <atom:updated>Fri, 31 Jul 2026 00:00:00 +0800</atom:updated>
      <dc:creator>熊鑫伟 (Xinwei Xiong)</dc:creator>
      <guid isPermaLink="true">https://cubxxw.com/zh/ai-agent/posts/use-auto-gpt/</guid>
      <description>本文重新审视 2023 年 AutoGPT Classic 的自治实验，说明它为何已停止维护、旧安装教程为何不再安全，并以 2026 年 AutoGPT Platform 为基线，给出从目标拆解、工作流分块、权限与凭据隔离、成本止损、人工审批到测试部署的完整迁移方法，帮助开发者保留实验价值，同时避开过时命令和生产风险。
</description>
      <category domain="category">Development</category>
      <category domain="tag">AI</category>
      <category domain="tag">Agent</category>
      <category domain="tag">Automation</category>
      <category domain="tag">LLM</category>
      <category domain="tag">Python</category>
      <category domain="tag">Docker</category>
      <category domain="tag">Open Source</category>
      <enclosure url="https://cubxxw.com/images/covers/ai-agent/2023/use-auto-gpt.png" type="image/jpeg" length="0" />
      <media:content url="https://cubxxw.com/images/covers/ai-agent/2023/use-auto-gpt.png" medium="image"><media:description>一辆发条小车穿过层层受控门框，象征 AutoGPT 从 Classic 自治实验走向可治理的工作流平台</media:description></media:content>
    </item>
  </channel>
</rss>
