https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/07-harness/04-deep-dive/技术 × 产品 × 机器人Hugo 0.160.1 & FixIt v0.4.6-20260517062719-ed8ec5c2zh-CNruohuyang@163.com (Cygnus Yang)ruohuyang@163.com (Cygnus Yang)Sat, 23 May 2026 00:00:00 +0800https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/07-harness/04-deep-dive/10-agent-loop/Sat, 23 May 2026 00:00:00 +0800ruohuyang@163.com (Cygnus Yang)https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/07-harness/04-deep-dive/10-agent-loop/<p>上一篇我们全景扫描了开源 Harness 生态。本章我们深入核心——Agent Loop，理解 observe→think→act→feedback 的持续循环。</p> <h2 class="heading-element" id="agent-loop-是什么"><span>Agent Loop 是什么</span> <a href="#agent-loop-%e6%98%af%e4%bb%80%e4%b9%88" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h2><p><strong>一句话定义</strong>：Agent 的核心心跳——持续循环的感知-推理-行动机制。</p>https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/07-harness/04-deep-dive/11-tool-system/Sat, 23 May 2026 00:00:00 +0800ruohuyang@163.com (Cygnus Yang)https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/07-harness/04-deep-dive/11-tool-system/<p>上一篇我们深入了 Agent Loop 的核心机制。本章我们看工具系统——Agent 的双手，如何从Bash扩展到MCP。</p> <h2 class="heading-element" id="工具系统是什么"><span>工具系统是什么</span> <a href="#%e5%b7%a5%e5%85%b7%e7%b3%bb%e7%bb%9f%e6%98%af%e4%bb%80%e4%b9%88" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h2><p><strong>一句话定义</strong>：Agent 可以采取的行动集合——文件读写、命令执行、网络请求、浏览器控制等。</p>https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/07-harness/04-deep-dive/12-permission-system/Sat, 23 May 2026 00:00:00 +0800ruohuyang@163.com (Cygnus Yang)https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/07-harness/04-deep-dive/12-permission-system/<p>上一篇我们剖析了工具系统的分发机制。本章我们看权限系统——保护安全的三级边界。</p> <h2 class="heading-element" id="权限系统是什么"><span>权限系统是什么</span> <a href="#%e6%9d%83%e9%99%90%e7%b3%bb%e7%bb%9f%e6%98%af%e4%bb%80%e4%b9%88" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h2><p><strong>一句话定义</strong>：安全边界和审批流程——控制哪些操作可以自动执行，哪些需要用户确认。</p>https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/07-harness/04-deep-dive/13-context-management/Sat, 23 May 2026 00:00:00 +0800ruohuyang@163.com (Cygnus Yang)https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/07-harness/04-deep-dive/13-context-management/<p>上一篇我们剖析了权限系统的三级边界。本章我们看上下文管理——防止溢出、压缩历史、惰性加载。</p> <h2 class="heading-element" id="上下文管理是什么"><span>上下文管理是什么</span> <a href="#%e4%b8%8a%e4%b8%8b%e6%96%87%e7%ae%a1%e7%90%86%e6%98%af%e4%bb%80%e4%b9%88" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h2><p><strong>一句话定义</strong>：管理 messages 数组的增长——防止溢出、压缩历史、按需加载知识。</p>https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/07-harness/04-deep-dive/14-mcp-integration/Sat, 23 May 2026 00:00:00 +0800ruohuyang@163.com (Cygnus Yang)https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/07-harness/04-deep-dive/14-mcp-integration/<p>上一篇我们剖析了上下文管理的四层策略。本章我们看 MCP集成——连接 AI 和外部系统的开放协议。</p> <h2 class="heading-element" id="mcp-是什么"><span>MCP 是什么</span> <a href="#mcp-%e6%98%af%e4%bb%80%e4%b9%88" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h2><p><strong>一句话定义</strong>：Model Context Protocol——连接 AI 和外部工具的开放协议，扩展 Harness 能力边界。</p>https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/07-harness/04-deep-dive/15-skills-system/Sat, 23 May 2026 00:00:00 +0800ruohuyang@163.com (Cygnus Yang)https://cygnusyang.github.io/posts/01-ai-%E5%B7%A5%E5%85%B7%E4%B8%8E%E6%99%BA%E8%83%BD%E4%BD%93/07-harness/04-deep-dive/15-skills-system/<p>上一篇我们剖析了 MCP 的扩展机制。本章我们看技能系统——按需注入领域知识的 Markdown 文件。</p> <h2 class="heading-element" id="技能系统是什么"><span>技能系统是什么</span> <a href="#%e6%8a%80%e8%83%bd%e7%b3%bb%e7%bb%9f%e6%98%af%e4%bb%80%e4%b9%88" class="heading-mark"> <svg class="octicon octicon-link" viewBox="0 0 16 16" version="1.1" width="16" height="16" aria-hidden="true"><path d="m7.775 3.275 1.25-1.25a3.5 3.5 0 1 1 4.95 4.95l-2.5 2.5a3.5 3.5 0 0 1-4.95 0 .751.751 0 0 1 .018-1.042.751.751 0 0 1 1.042-.018 1.998 1.998 0 0 0 2.83 0l2.5-2.5a2.002 2.002 0 0 0-2.83-2.83l-1.25 1.25a.751.751 0 0 1-1.042-.018.751.751 0 0 1-.018-1.042Zm-4.69 9.64a1.998 1.998 0 0 0 2.83 0l1.25-1.25a.751.751 0 0 1 1.042.018.751.751 0 0 1 .018 1.042l-1.25 1.25a3.5 3.5 0 1 1-4.95-4.95l2.5-2.5a3.5 3.5 0 0 1 4.95 0 .751.751 0 0 1-.018 1.042.751.751 0 0 1-1.042.018 1.998 1.998 0 0 0-2.83 0l-2.5 2.5a1.998 1.998 0 0 0 0 2.83Z"></path></svg> </a> </h2><p><strong>一句话定义</strong>：按需注入的领域知识 Markdown 文件——技术文档、编码规范、领域知识。</p>