<?xml version="1.0" encoding="UTF-8"?><?xml-stylesheet type="text/xsl" href="/rss.xsl" ?>

<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/">
  <channel>
    <title>面条实验室</title>
    <link>https://miantiao.me</link>
    <description>折腾些什么玩意</description>
    <atom:link href="https://miantiao.me/diyfeed.xml" rel="self" type="application/rss+xml" />
    <lastBuildDate>Tue, 17 Nov 2026 12:22:30 GMT</lastBuildDate>
    <item>
      <title>用 AI Agent 搞了个全自动 AI 周刊</title>
      <link>https://miantiao.me/posts/aigc-weekly/</link>
      <guid>https://miantiao.me/posts/aigc-weekly/</guid>
      <description>一个由 Claude Agent 全自动驱动的 AIGC 周刊，基于 Next.js 15 和 Cloudflare 全栈构建，每周自动搜集、筛选、生成并发布 AI 领域的精选内容。</description>
      <pubDate>Sat, 13 Dec 2025 05:10:31 GMT</pubDate>
      <content:encoded><![CDATA[<link rel="preload" as="image" href="https://aigc-weekly.agi.li/opengraph-image.png"/><p>最近我做了一个有点意思的项目——<strong>AIGC Weekly</strong>。这不仅是一份关注 AIGC（人工智能生成内容）领域的周刊，更是一次关于 <strong>Agentic AI（代理智能）</strong> 的深度实验。</p>
<p>今天想和大家聊聊，我为什么要做这个项目，以及它是如何通过一群 AI Agent 自动运转起来的。</p>
<p><button type="button" data-zoomable="" aria-label="放大图片：AIGC Weekly 周刊封面"><img src="https://aigc-weekly.agi.li/opengraph-image.png" alt="AIGC Weekly 周刊封面"/></button></p>
<h2>为什么要做 AIGC Weekly？</h2>
<p>作为一名开发者，我每天都会被海量的 AI 资讯淹没：</p>
<ul>
<li>GitHub 上每天冒出无数个新项目；</li>
<li>Twitter 和 Hacker News 上的讨论层出不穷；</li>
<li>各种技术博客更新得比我读得还快。</li>
</ul>
<p><strong>信息过载</strong>是这个时代的通病，尤其是在 AI 这样一个日新月异的领域。我希望能有一个工具，能帮我从这些噪音中筛选出真正有价值的信号——那些高质量的开源项目、深度的技术观点和实用的工具发现。</p>
<p>但我不想人工来做这件事。</p>
<p>既然我们已经有了像 Claude 这样强大的 LLM，为什么不让 AI 自己来当编辑呢？于是，<strong>AIGC Weekly</strong> 诞生了。它的目标很简单：<strong>全自动化、高质量、无需人工干预</strong>。</p>
<h2>AI Agent 如何运转一份周刊</h2>
<p>AIGC Weekly 的核心不仅是一个网站，而是一套<strong>自主运行的流水线</strong>。每周日早上八点，当大部分人还在休息时，我的 AI Agent 们就开始上班了。</p>
<p>整个流程完全自动化，大概分为这几步：</p>
<figure class="pos_relative min-w_0 bd_1px_dashed_token(colors.line-strong) p_1lh_2ch [&amp;_&gt;_figcaption]:pos_relative [&amp;_&gt;_figcaption]:w_fit-content [&amp;_&gt;_figcaption]:max-w_100% [&amp;_&gt;_figcaption]:m_-1.5lh_auto_0.5lh [&amp;_&gt;_figcaption]:px_1ch [&amp;_&gt;_figcaption]:bg_bg [&amp;_&gt;_figcaption]:c_accent [&amp;_&gt;_figcaption]:ta_center [&amp;_&gt;_figcaption]:ov-wrap_anywhere [&amp;_&gt;_div_&gt;_:first-child]:mt_0 [&amp;_&gt;_div_&gt;_:last-child]:mb_0"><figcaption><span aria-hidden="true">[ </span>AI 周刊自动化流水线<span aria-hidden="true"> ]</span></figcaption><div><div class="[&amp;_&gt;_ol]:m_0 [&amp;_&gt;_ol]:pl_3ch [&amp;_&gt;_ol]:li-t_decimal-leading-zero [&amp;_&gt;_ol_&gt;_li]:pl_1ch [&amp;_&gt;_ol_&gt;_li::marker]:c_accent [&amp;_&gt;_ol_&gt;_li::marker]:fv-num_tabular-nums [&amp;_&gt;_ol_&gt;_li_+_li]:mt_1lh [&amp;_&gt;_ol_&gt;_li_&gt;_p_+_p]:mt_0.5lh"><ol>
<li>
<p>自动搜集（Research）</p>
<p>Agent 会从 Hacker News、知名 AI 博客、Telegram 频道等 15+ 个信源抓取最新的 AI 资讯。</p>
</li>
<li>
<p>智能筛选（Curate）</p>
<p>这不是简单的关键词匹配。Agent 会像一个挑剔的主编，根据 GitHub Star 数增长趋势、社区讨论热度等「智能标准」来评估每一条内容的价值，只有真正优质的内容才能通过筛选。</p>
</li>
<li>
<p>内容生成（Write）</p>
<p>筛选出的素材会被交给负责写作的 Agent。它会撰写标题、摘要，并整理成一篇结构清晰的周刊文章。</p>
</li>
<li>
<p>自动发布（Publish）</p>
<p>最后，Agent 会通过 <a href="https://modelcontextprotocol.io/" rel="nofollow noopener noreferrer" target="_blank">MCP（Model Context Protocol）</a> 直接与 CMS 交互，将文章发布到网站上，并同步更新 RSS 源。</p>
</li>
</ol></div></div></figure>
<p>这一切，都在 Cloudflare 的云端容器里静悄悄地完成。</p>
<h2>技术栈：Serverless + AI Agent</h2>
<p>作为一个技术人，我也想借此机会实践一下最新的技术栈。AIGC Weekly 是一个典型的 <strong>Modern Web + AI</strong> 应用：</p>
<figure class="pos_relative min-w_0 bd_1px_dashed_token(colors.line-strong) p_1lh_2ch [&amp;_&gt;_figcaption]:pos_relative [&amp;_&gt;_figcaption]:w_fit-content [&amp;_&gt;_figcaption]:max-w_100% [&amp;_&gt;_figcaption]:m_-1.5lh_auto_0.5lh [&amp;_&gt;_figcaption]:px_1ch [&amp;_&gt;_figcaption]:bg_bg [&amp;_&gt;_figcaption]:c_accent [&amp;_&gt;_figcaption]:ta_center [&amp;_&gt;_figcaption]:ov-wrap_anywhere [&amp;_&gt;_div_&gt;_:first-child]:mt_0 [&amp;_&gt;_div_&gt;_:last-child]:mb_0"><div><div role="region" aria-label="表格" tabindex="0" class="min-w_0 ov-x_auto focusVisible:ring_2px_solid_token(colors.accent) focusVisible:ring-o_0.5ch [&amp;_table]:d_table [&amp;_table]:m_0 [&amp;_table]:w_100% [&amp;_table]:bd-cl_separate [&amp;_table]:bd-sp_0 [&amp;_th,_&amp;_td]:min-w_10ch [&amp;_th,_&amp;_td]:p_0.5lh_1ch [&amp;_th,_&amp;_td]:bd_0 [&amp;_th,_&amp;_td]:bd-s_1px_dashed_token(colors.line) [&amp;_th,_&amp;_td]:fv-num_tabular-nums [&amp;_th,_&amp;_td]:ov-wrap_anywhere [&amp;_tr_&gt;_:first-child]:bd-s_0 [&amp;_tr_&gt;_:first-child]:ps_0 [&amp;_tr_&gt;_:last-child]:pe_0 [&amp;_thead_th]:pt_0 [&amp;_thead_th]:c_accent [&amp;_thead_th]:white-space_nowrap [&amp;_thead_th]:bd-b_1px_dashed_token(colors.line-strong) [&amp;_tbody_th]:fw_normal [&amp;_tbody_tr:last-child_&gt;_*]:pb_0"><table><thead><tr><th scope="col">组件</th><th scope="col">技术选型</th><th scope="col">说明</th></tr></thead><tbody><tr><td>前端框架</td><td>Next.js 15（App Router）+ OpenNext</td><td>保证极致的加载速度和 SEO 友好</td></tr><tr><td>内容管理</td><td>Payload CMS 3.0</td><td>Headless CMS，灵活强大，适合与 AI 进行 API 级别的交互</td></tr><tr><td>数据库</td><td>Cloudflare D1（SQLite）</td><td>轻量且快</td></tr><tr><td>图片存储</td><td>Cloudflare R2</td><td></td></tr><tr><td>Agent 运行环境</td><td>Cloudflare Containers</td><td></td></tr><tr><td>AI 核心</td><td>Anthropic Claude Agent SDK</td><td>系统的大脑，负责理解任务、规划流程和执行操作</td></tr><tr><td>数据抓取</td><td>Firecrawl</td><td>高效地从网页中提取结构化数据</td></tr></tbody></table></div></div></figure>
<p>这套架构最大的好处是 <strong>Serverless</strong>。我不需要维护复杂的服务器，系统会根据负载自动伸缩，而且成本极低。</p>
<h2>什么是 Slow is Fast？</h2>
<p>在开发这个项目的过程中，我一直遵循 <strong>「Slow is Fast」</strong> 的理念。</p>
<p>在这个 AI 狂飙的时代，我们很容易陷入追求「快」的焦虑中：快速上线、快速迭代、快速变现。但我认为，对于一个信息类产品，<strong>质量远比速度重要</strong>。</p>
<ul>
<li><strong>推理质量</strong>：我宁愿让 Agent 多花几分钟去交叉验证信息的准确性，去打磨一段摘要的措辞，也不希望它秒出一篇充满幻觉的垃圾文。</li>
<li><strong>架构设计</strong>：花时间把 Agent 的 Workflow 设计好，把 CMS 的数据结构定义清晰，这在初期看起来很慢，但长远来看，它让系统的维护和扩展变得异常简单。</li>
</ul>
<p>AIGC Weekly 不追求实时更新，而是选择周更。这本身就是一种态度：<strong>在快节奏的洪流中，沉淀下来的才是精华。</strong></p>
<h2>开始探索 AIGC Weekly</h2>
<p>AIGC Weekly 目前还在早期阶段，但它已经能稳定地为我（以及现在的许多订阅者）提供服务了。</p>
<p>这是一个完全开源的项目。如果你对 <strong>AI Agent 开发</strong>、<strong>Next.js 实战</strong>或者 <strong>Serverless 架构</strong>感兴趣，欢迎来 GitHub 看看代码，或者直接订阅周刊，看看 AI 挑选的内容是否合你口味。</p>
<ul>
<li>周刊地址：<a href="https://aigc-weekly.agi.li/" rel="nofollow noopener noreferrer" target="_blank">aigc-weekly.agi.li</a></li>
<li>GitHub 仓库：<a href="https://github.com/miantiao-me/aigc-weekly" rel="nofollow noopener noreferrer" target="_blank">miantiao-me/aigc-weekly</a></li>
</ul>
<p>期待在这个由 AI 驱动的小小角落里，与你相遇。</p><img src="https://c.statcounter.com/9823304/0/92a1d06c/1/" alt="" width="1" height="1" referrerPolicy="no-referrer-when-downgrade" />]]></content:encoded>
    </item>
    <item>
      <title>hink - 只需 10 行代码的短链接系统</title>
      <link>https://miantiao.me/posts/hink/</link>
      <guid>https://miantiao.me/posts/hink/</guid>
      <description>一个用 Git 和 Serverless 平台实现的极简短链接系统，代码量不到 10 行。</description>
      <pubDate>Sun, 31 Aug 2025 04:50:55 GMT</pubDate>
      <content:encoded><![CDATA[<link rel="preload" as="image" href="https://static.miantiao.me/share/2025/ePMX6q/Zsb50p.png"/><p>和大家分享一个最近做的小工具——<a href="https://github.com/miantiao-me/hink" rel="nofollow noopener noreferrer" target="_blank"><strong>hink</strong></a>，一个用不到 10 行代码实现的短链接系统。</p>
<p>这个工具利用 Git 和 Serverless 平台，实现了短链接生成和访问统计功能。</p>
<h2>核心原理：Git 提交哈希值的妙用</h2>
<p>hink 的核心思路很简单：用 Git 的空提交哈希值作为短链接的唯一标识符，把原始长链接存储在提交信息中。访问短链接时，系统通过 GitHub 的 <code>.patch</code> 文件接口读取提交信息，提取长链接并重定向。结合云平台的 WAF（Web 应用防火墙）分析面板，还能实现访问统计。</p>
<p>目前，这个方案已经在以下平台测试通过：</p>
<ul>
<li>Cloudflare Workers/Snippets + Cloudflare WAF (Pro)</li>
<li>腾讯云 EdgeOne (Free)</li>
<li>阿里云 ESA (Free)</li>
</ul>
<h2>代码：极简实现</h2>
<p>hink 的代码非常短小，核心逻辑只有几行。以下是针对不同平台的实现。</p>
<h3>Cloudflare Workers / 阿里云 ESA 版本</h3>
<pre class="kirari"><code class="language-js"><span class="kirari-keyword">const</span> <span class="kirari-variable">GIT_REPO</span> <span class="kirari-operator">=</span> <span class="kirari-string">&quot;https://github.com/miantiao-me/hink&quot;</span>
<span class="kirari-keyword">export</span> <span class="kirari-keyword">default</span> <span class="kirari-punctuation">{</span>
  <span class="kirari-keyword">async</span> <span class="kirari-function">fetch</span><span class="kirari-punctuation">(</span><span class="kirari-variable">request</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-keyword">const</span> <span class="kirari-punctuation">{</span> <span class="kirari-variable">pathname</span> <span class="kirari-punctuation">}</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">new</span> <span class="kirari-function">URL</span><span class="kirari-punctuation">(</span><span class="kirari-variable">request</span><span class="kirari-punctuation">.</span><span class="kirari-property">url</span><span class="kirari-punctuation">)</span>
    <span class="kirari-keyword">const</span> <span class="kirari-variable">gitPatch</span> <span class="kirari-operator">=</span> <span class="kirari-string">`</span><span class="kirari-punctuation">${</span><span class="kirari-variable">GIT_REPO</span><span class="kirari-punctuation">}</span><span class="kirari-string">/commit</span><span class="kirari-punctuation">${</span><span class="kirari-variable">pathname</span><span class="kirari-punctuation">}</span><span class="kirari-string">.patch`</span>
    <span class="kirari-keyword">const</span> <span class="kirari-variable">patch</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">await</span> <span class="kirari-function">fetch</span><span class="kirari-punctuation">(</span><span class="kirari-variable">gitPatch</span><span class="kirari-punctuation">,</span> <span class="kirari-punctuation">{</span> <span class="kirari-property">cf</span><span class="kirari-punctuation">:</span> <span class="kirari-punctuation">{</span> <span class="kirari-property">cacheEverything</span><span class="kirari-punctuation">:</span> <span class="kirari-boolean">true</span><span class="kirari-punctuation">,</span> <span class="kirari-property">cacheTtlByStatus</span><span class="kirari-punctuation">:</span> <span class="kirari-punctuation">{</span> <span class="kirari-string">&#x27;200-299&#x27;</span><span class="kirari-punctuation">:</span> <span class="kirari-number">86400</span> <span class="kirari-punctuation">}</span> <span class="kirari-punctuation">}</span><span class="kirari-punctuation">}</span><span class="kirari-punctuation">)</span><span class="kirari-punctuation">.</span><span class="kirari-property">then</span><span class="kirari-punctuation">(</span><span class="kirari-variable">res</span> <span class="kirari-operator">=&gt;</span> <span class="kirari-variable">res</span><span class="kirari-punctuation">.</span><span class="kirari-property">text</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span><span class="kirari-punctuation">)</span>
    <span class="kirari-keyword">const</span> <span class="kirari-variable">url</span> <span class="kirari-operator">=</span> <span class="kirari-variable">pathname</span> <span class="kirari-operator">===</span> <span class="kirari-string">&#x27;/&#x27;</span> <span class="kirari-operator">?</span> <span class="kirari-variable">GIT_REPO</span> <span class="kirari-punctuation">:</span> <span class="kirari-variable">patch</span><span class="kirari-punctuation">.</span><span class="kirari-property">match</span><span class="kirari-punctuation">(</span><span class="kirari-regex">/^Subject:\s*\[PATCH\](.*)$/m</span><span class="kirari-punctuation">)</span><span class="kirari-operator">?.</span><span class="kirari-punctuation">[</span><span class="kirari-number">1</span><span class="kirari-punctuation">]</span><span class="kirari-operator">?.</span><span class="kirari-property">trim</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span>
    <span class="kirari-keyword">return</span> <span class="kirari-variable">Response</span><span class="kirari-punctuation">.</span><span class="kirari-property">redirect</span><span class="kirari-punctuation">(</span><span class="kirari-variable">url</span> <span class="kirari-operator">||</span> <span class="kirari-variable">GIT_REPO</span><span class="kirari-punctuation">)</span>
  <span class="kirari-punctuation">}</span>
<span class="kirari-punctuation">}</span>
</code></pre>
<h3>腾讯云 EdgeOne 版本</h3>
<pre class="kirari"><code class="language-js"><span class="kirari-keyword">const</span> <span class="kirari-variable">GIT_REPO</span> <span class="kirari-operator">=</span> <span class="kirari-string">&quot;https://github.com/miantiao-me/hink&quot;</span>
<span class="kirari-function">addEventListener</span><span class="kirari-punctuation">(</span><span class="kirari-string">&quot;fetch&quot;</span><span class="kirari-punctuation">,</span> <span class="kirari-keyword">async</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">event</span><span class="kirari-punctuation">)</span> <span class="kirari-operator">=&gt;</span> <span class="kirari-punctuation">{</span>
  <span class="kirari-keyword">const</span> <span class="kirari-punctuation">{</span> <span class="kirari-variable">pathname</span> <span class="kirari-punctuation">}</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">new</span> <span class="kirari-function">URL</span><span class="kirari-punctuation">(</span><span class="kirari-variable">event</span><span class="kirari-punctuation">.</span><span class="kirari-property">request</span><span class="kirari-punctuation">.</span><span class="kirari-property">url</span><span class="kirari-punctuation">)</span>
  <span class="kirari-keyword">const</span> <span class="kirari-variable">gitPatch</span> <span class="kirari-operator">=</span> <span class="kirari-string">`</span><span class="kirari-punctuation">${</span><span class="kirari-variable">GIT_REPO</span><span class="kirari-punctuation">}</span><span class="kirari-string">/commit</span><span class="kirari-punctuation">${</span><span class="kirari-variable">pathname</span><span class="kirari-punctuation">}</span><span class="kirari-string">.patch`</span>
  <span class="kirari-keyword">const</span> <span class="kirari-variable">patch</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">await</span> <span class="kirari-function">fetch</span><span class="kirari-punctuation">(</span><span class="kirari-variable">gitPatch</span><span class="kirari-punctuation">)</span><span class="kirari-punctuation">.</span><span class="kirari-property">then</span><span class="kirari-punctuation">(</span><span class="kirari-variable">res</span> <span class="kirari-operator">=&gt;</span> <span class="kirari-variable">res</span><span class="kirari-punctuation">.</span><span class="kirari-property">text</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span><span class="kirari-punctuation">)</span>
  <span class="kirari-keyword">const</span> <span class="kirari-variable">url</span> <span class="kirari-operator">=</span> <span class="kirari-variable">pathname</span> <span class="kirari-operator">===</span> <span class="kirari-string">&#x27;/&#x27;</span> <span class="kirari-operator">?</span> <span class="kirari-variable">GIT_REPO</span> <span class="kirari-punctuation">:</span> <span class="kirari-variable">patch</span><span class="kirari-punctuation">.</span><span class="kirari-property">match</span><span class="kirari-punctuation">(</span><span class="kirari-regex">/^Subject:\s*\[PATCH\](.*)$/m</span><span class="kirari-punctuation">)</span><span class="kirari-operator">?.</span><span class="kirari-punctuation">[</span><span class="kirari-number">1</span><span class="kirari-punctuation">]</span><span class="kirari-operator">?.</span><span class="kirari-property">trim</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span>
  <span class="kirari-variable">event</span><span class="kirari-punctuation">.</span><span class="kirari-property">respondWith</span><span class="kirari-punctuation">(</span><span class="kirari-keyword">new</span> <span class="kirari-function">Response</span><span class="kirari-punctuation">(</span><span class="kirari-null">null</span><span class="kirari-punctuation">,</span> <span class="kirari-punctuation">{</span> <span class="kirari-property">status</span><span class="kirari-punctuation">:</span> <span class="kirari-number">302</span><span class="kirari-punctuation">,</span> <span class="kirari-property">headers</span><span class="kirari-punctuation">:</span> <span class="kirari-punctuation">{</span> <span class="kirari-property">Location</span><span class="kirari-punctuation">:</span> <span class="kirari-variable">url</span> <span class="kirari-operator">||</span> <span class="kirari-variable">GIT_REPO</span> <span class="kirari-punctuation">}</span> <span class="kirari-punctuation">}</span><span class="kirari-punctuation">)</span><span class="kirari-punctuation">)</span>
<span class="kirari-punctuation">}</span><span class="kirari-punctuation">)</span><span class="kirari-punctuation">;</span>
</code></pre>
<p>部署到 Serverless 平台后，绑定一个域名，就能拥有自己的短链接服务。</p>
<h2>为什么做 hink？</h2>
<p>短链接服务并不新鲜，市面上有很多工具。但 hink 的目标是探索一种极简且有趣的实现方式。利用 Git 提交哈希值，避免了数据库管理，借助 GitHub 实现存储，而云平台的 WAF 功能让访问统计变得简单。这对我来说是一次技术实验，用最少的代码解决实际问题。</p>
<h2>演示效果：三大平台运行情况</h2>
<p>我在 Cloudflare Workers、阿里云 ESA 和腾讯云 EdgeOne 上部署了 hink，并通过 WAF 面板查看访问统计。以下是各平台的运行截图：</p>
<h3>Cloudflare Workers</h3>
<p><button type="button" data-zoomable="" aria-label="放大图片：Cloudflare"><img src="https://static.miantiao.me/share/2025/ePMX6q/Zsb50p.png" alt="Cloudflare"/></button></p>
<h3>阿里云 ESA</h3>
<p><button type="button" data-zoomable="" aria-label="放大图片：Alibaba"><img loading="lazy" decoding="async" src="https://static.miantiao.me/share/2025/JiOSmY/6oDC36.png" alt="Alibaba"/></button></p>
<h3>腾讯云 EdgeOne</h3>
<p><button type="button" data-zoomable="" aria-label="放大图片：Tencent"><img loading="lazy" decoding="async" src="https://static.miantiao.me/share/2025/dyzFzs/efOSKl.png" alt="Tencent"/></button></p>
<h2>项目地址</h2>
<p><a href="https://github.com/miantiao-me/hink" rel="nofollow noopener noreferrer" target="_blank"><img loading="lazy" decoding="async" src="https://github.html.zone/miantiao-me/hink" alt="miantiao-me/hink - GitHub"/></a></p><img src="https://c.statcounter.com/9823304/0/92a1d06c/1/" alt="" width="1" height="1" referrerPolicy="no-referrer-when-downgrade" />]]></content:encoded>
    </item>
    <item>
      <title>解密 SSH AI Chat：它是如何运行的</title>
      <link>https://miantiao.me/posts/how-ssh-ai-chat-works/</link>
      <guid>https://miantiao.me/posts/how-ssh-ai-chat-works/</guid>
      <description>深入探索 SSH AI Chat 的技术架构与实现原理，了解如何通过 SSH 协议构建创新的 AI 聊天应用，以及终端 UI、认证系统和流式响应等核心技术的巧妙设计。</description>
      <pubDate>Fri, 01 Aug 2025 14:07:03 GMT</pubDate>
      <content:encoded><![CDATA[<p>大家好，我是面条，今天想和大家分享我最近的一个项目——<strong>SSH AI Chat</strong>。</p>
<h2>项目简介</h2>
<p>SSH AI Chat 是一个可以通过 SSH 直接连接的 AI 聊天应用。使用方式非常简单：</p>
<pre class="kirari"><code class="language-bash"><span class="kirari-variable">ssh</span> <span class="kirari-variable">username</span>@<span class="kirari-variable">chat</span>.<span class="kirari-variable">agi</span>.<span class="kirari-variable">li</span>
</code></pre>
<p>没错，就这么简单！你不需要安装任何客户端，不需要打开浏览器，只需要一个 SSH 客户端，就能和 AI 进行对话。</p>
<p>作为一个对 TUI 应用有着浓厚兴趣的开发者，我一直觉得在终端里聊天是一件很酷的事情。其实我最初是被 itter.sh 这个网站惊艳到了——居然能用 SSH 访问社交网络！这让我意识到，原来 SSH 不只是用来连服务器的，还可以做很多有趣的事情。</p>
<p>于是就有了这个想法：如果能用 SSH 和 AI 聊天，那该多酷啊！不需要安装任何软件，不需要打开浏览器，只要在终端里输入 <code>ssh yourname@chat.agi.li</code> 就能开始对话。</p>
<h2>项目架构</h2>
<h3>核心技术栈</h3>
<ul>
<li><strong>SSH 服务器</strong>：使用 Node.js + ssh2 模块</li>
<li><strong>UI 框架</strong>：React + Ink（用于终端渲染）</li>
<li><strong>数据库</strong>：PostgreSQL / PGLite（可选）</li>
<li><strong>缓存</strong>：Redis / ioredis-mock（可选）</li>
<li><strong>AI 集成</strong>：Vercel AI SDK</li>
</ul>
<h3>架构特点</h3>
<ul>
<li>无需安装客户端</li>
<li>跨平台终端支持</li>
<li>GitHub 公钥认证</li>
<li>实时流式响应</li>
<li>多模型 AI 支持</li>
</ul>
<h3>系统架构图</h3>
<figure class="pos_relative min-w_0 bd_1px_dashed_token(colors.line-strong) p_1lh_2ch [&amp;_&gt;_figcaption]:pos_relative [&amp;_&gt;_figcaption]:w_fit-content [&amp;_&gt;_figcaption]:max-w_100% [&amp;_&gt;_figcaption]:m_-1.5lh_auto_0.5lh [&amp;_&gt;_figcaption]:px_1ch [&amp;_&gt;_figcaption]:bg_bg [&amp;_&gt;_figcaption]:c_accent [&amp;_&gt;_figcaption]:ta_center [&amp;_&gt;_figcaption]:ov-wrap_anywhere [&amp;_&gt;_div_&gt;_:first-child]:mt_0 [&amp;_&gt;_div_&gt;_:last-child]:mb_0"><figcaption><span aria-hidden="true">[ </span>SSH AI Chat 系统架构<span aria-hidden="true"> ]</span></figcaption><div><div class="d_grid gap_1lh"><p class="d_flex flex-wrap_wrap ai_baseline gap_0.5lh_1ch [&amp;_strong]:c_accent"><span class="bd_1px_solid_token(colors.line) px_1ch min-w_0 ov-wrap_anywhere">SSH Client（<code>ssh username@chat.agi.li</code>）</span><span> → </span><span class="bd_1px_solid_token(colors.line) px_1ch min-w_0 ov-wrap_anywhere"><strong>SSH Server（Node.js + ssh2）</strong></span><span> → </span><span class="bd_1px_solid_token(colors.line) px_1ch min-w_0 ov-wrap_anywhere">React App（Ink UI + React Hooks）</span></p><p class="d_flex flex-wrap_wrap ai_baseline gap_0.5lh_1ch [&amp;_strong]:c_accent"><span class="bd_1px_solid_token(colors.line) px_1ch min-w_0 ov-wrap_anywhere">SSH Server</span><span> → </span><span class="bd_1px_solid_token(colors.line) px_1ch min-w_0 ov-wrap_anywhere">AI Services（OpenAI API / Gemini API / DeepSeek API）</span></p></div></div></figure>
<h2>核心模块解析</h2>
<h3>1. SSH 服务器模块</h3>
<p>这是整个应用的核心，负责处理 SSH 连接和认证。系统会自动处理密钥验证、GitHub 公钥认证、登录限制和速率限制。</p>
<h3>2. 认证系统</h3>
<p>最巧妙的设计是使用 GitHub 公钥认证。用户不需要注册，直接使用 GitHub SSH 密钥就能登录。系统会获取用户的 GitHub 公钥进行验证，每 6 小时缓存一次，既安全又高效。</p>
<h3>3. 终端 UI 系统</h3>
<p>使用 Ink 框架在终端中渲染 React 组件。想象一下，你平时写的 React 组件，现在不是在浏览器里渲染，而是在终端里显示！支持多语言界面、实时聊天、历史记录、模型选择和响应式布局。</p>
<h3>4. 聊天系统</h3>
<p>使用 Vercel AI SDK 处理 AI 对话。当你在终端里输入消息时，系统会接收消息、加载历史对话、选择模型、实时显示流式响应，并保存对话记录。支持流式响应、多模型支持、思维链展示和对话历史管理。</p>
<h2>技术难点和解决方案</h2>
<h3>1. 终端渲染的挑战</h3>
<p>最大的挑战是在终端中实现复杂的 UI 界面。使用 Ink 框架将 React 组件渲染到终端，实现虚拟 PTY 来处理终端 I/O。在终端里显示 AI 回复的 Markdown 需要用专门的 worker 进程处理转换，才能显示粗体、斜体、代码块。</p>
<h3>2. SSH 会话管理</h3>
<p>管理多个 SSH 会话和状态需要为每个连接创建独立的 React 应用实例，使用 Context API 管理全局状态，并实现会话生命周期管理。</p>
<h3>3. 实时流式响应</h3>
<p>AI 的回复是流式的，如果每收到一个字节就刷新界面，终端会直接卡死。使用 Vercel AI SDK 的 <code>streamText</code> 配合节流更新，<strong>每 300ms 更新一次</strong>，既流畅又不卡。</p>
<h3>4. 数据存储的灵活性</h3>
<p>项目支持 PostgreSQL 和 PGLite 两种数据库，以及 Redis 和内存缓存，让项目既能独立运行又能部署到生产环境。</p>
<h2>有趣的设计细节</h2>
<h3>1. GitHub 认证</h3>
<p>最酷的设计！用户不需要注册，直接使用 GitHub SSH 密钥就能登录，既方便又安全。</p>
<h3>2. 多模型支持</h3>
<p>支持 DeepSeek-V3/DeepSeek-R1、Gemini-2.5-Flash/Gemini-2.5-Pro 等多个 AI 模型，包括思维链展示。需要处理不同模型 API 的差异，并提供统一的接口。</p>
<h3>3. 国际化支持</h3>
<p>完整的 i18n 支持，通过 <code>LANG</code> 环境变量自动检测用户语言偏好，支持中英文切换。</p>
<h3>4. 键盘快捷键</h3>
<p><code>Ctrl+C</code> 退出应用，<code>N</code> 新建对话，<code>I</code> 聚焦输入框，<code>?</code> 查看帮助。还有一些小彩蛋功能。</p>
<h2>开发心得</h2>
<h3>1. 终端应用的可能性</h3>
<p>这个项目让我看到了终端应用的巨大潜力。通过 Ink 框架，我们可以在终端中实现复杂的交互界面。</p>
<h3>2. SSH 的妙用</h3>
<p>SSH 不仅仅是一个远程管理工具，它还是一个强大的应用平台。通过 SSH，我们可以实现跨平台的客户端应用，用户不需要安装任何额外的软件。</p>
<h3>3. 现代化的技术栈</h3>
<p>虽然这是一个终端应用，但我们使用了最现代的技术栈：React、TypeScript、Vercel AI SDK 等。这证明了终端应用也可以很「现代化」。</p>
<h3>4. 遇到的坑</h3>
<p>SSH 终端兼容性问题、流式输出的性能问题、历史对话管理的性能挑战，这些都是开发过程中需要克服的难点。</p>
<h2>未来展望</h2>
<p>计划支持更多的 AI 模型，包括 <a href="https://ollama.com/" rel="nofollow noopener noreferrer" target="_blank">Ollama</a> 等本地模型，以及支持 <a href="https://modelcontextprotocol.io/" rel="nofollow noopener noreferrer" target="_blank">MCP（Model Context Protocol）</a>协议，让用户可以通过插件扩展功能。</p>
<h2>总结</h2>
<p>SSH AI Chat 是一个融合了多种技术的创新项目。它展示了：</p>
<ul>
<li>终端应用的现代化可能性</li>
<li>SSH 协议的灵活应用</li>
<li>React 在不同平台上的适应性</li>
<li>AI 技术的普及化</li>
</ul>
<p>这个项目让我意识到，技术不只是为了解决问题，也可以很有趣。把 SSH 和 AI 结合起来，创造出了意想不到的体验。</p>
<p>希望这个项目能给大家带来一些启发，让我们一起探索技术的边界！</p>
<h2>体验地址</h2>
<p>想要体验的话，把用户名换成你的 GitHub 用户名，直接执行：</p>
<pre class="kirari"><code class="language-bash"><span class="kirari-variable">ssh</span> <span class="kirari-variable">your-github-username</span>@<span class="kirari-variable">chat</span>.<span class="kirari-variable">agi</span>.<span class="kirari-variable">li</span>
</code></pre>
<p>项目地址：<a href="https://github.com/miantiao-me/ssh-ai-chat" rel="nofollow noopener noreferrer" target="_blank">miantiao-me/ssh-ai-chat</a></p>
<hr/>
<p>如果你对这个项目有任何问题或建议，欢迎在 GitHub 上讨论。也欢迎 Star 这个项目，你的支持是我继续开发的动力！</p><img src="https://c.statcounter.com/9823304/0/92a1d06c/1/" alt="" width="1" height="1" referrerPolicy="no-referrer-when-downgrade" />]]></content:encoded>
    </item>
    <item>
      <title>在 Docker 沙箱中运行 MCP Server</title>
      <link>https://miantiao.me/posts/guide-to-running-mcp-server-in-a-sandbox/</link>
      <guid>https://miantiao.me/posts/guide-to-running-mcp-server-in-a-sandbox/</guid>
      <description>利用 Docker 运行 MCP Server 并转换为 MCP SSE 协议，减少 npx 和 uvx 直接运行带来的任意文件读取风险。</description>
      <pubDate>Wed, 23 Apr 2025 14:01:48 GMT</pubDate>
      <content:encoded><![CDATA[<link rel="preload" as="image" href="https://static.miantiao.me/share/ZUy0MY/twitter.jpeg"/><p>MCP 是今年 AI 开发行业很热门的一个协议，但是由于它的 C/S 架构，导致使用者必须在本地运行 MCP Server。</p>
<p>MCP Server 常见的运行方式有 npx(NPM 生态)、uvx(Python 生态)、Docker 等 stdio 方式和 HTTP(SSE/Streaming) 方式。但是 npx 和 uvx 运行命令有着极大的风险。如果不慎执行恶意软件包，可能会导致隐私数据泄露带来极大的安全风险。具体可以看 Invariant 的 <a href="https://invariantlabs.ai/blog/mcp-security-notification-tool-poisoning-attacks" rel="nofollow noopener noreferrer" target="_blank">MCP Security Notification: Tool Poisoning Attacks</a> 这篇文章。</p>
<p>作为一个软件行业从业者，对安全的关注度极高。让 ChatGPT 整理了一下最近 5 年的 NPM 和 PyPI 供应链攻击事件，让人不寒而栗。</p>
<figure class="pos_relative min-w_0 bd_1px_dashed_token(colors.line-strong) p_1lh_2ch [&amp;_&gt;_figcaption]:pos_relative [&amp;_&gt;_figcaption]:w_fit-content [&amp;_&gt;_figcaption]:max-w_100% [&amp;_&gt;_figcaption]:m_-1.5lh_auto_0.5lh [&amp;_&gt;_figcaption]:px_1ch [&amp;_&gt;_figcaption]:bg_bg [&amp;_&gt;_figcaption]:c_accent [&amp;_&gt;_figcaption]:ta_center [&amp;_&gt;_figcaption]:ov-wrap_anywhere [&amp;_&gt;_div_&gt;_:first-child]:mt_0 [&amp;_&gt;_div_&gt;_:last-child]:mb_0"><div><div role="region" aria-label="表格" tabindex="0" class="min-w_0 ov-x_auto focusVisible:ring_2px_solid_token(colors.accent) focusVisible:ring-o_0.5ch [&amp;_table]:d_table [&amp;_table]:m_0 [&amp;_table]:w_100% [&amp;_table]:bd-cl_separate [&amp;_table]:bd-sp_0 [&amp;_th,_&amp;_td]:min-w_10ch [&amp;_th,_&amp;_td]:p_0.5lh_1ch [&amp;_th,_&amp;_td]:bd_0 [&amp;_th,_&amp;_td]:bd-s_1px_dashed_token(colors.line) [&amp;_th,_&amp;_td]:fv-num_tabular-nums [&amp;_th,_&amp;_td]:ov-wrap_anywhere [&amp;_tr_&gt;_:first-child]:bd-s_0 [&amp;_tr_&gt;_:first-child]:ps_0 [&amp;_tr_&gt;_:last-child]:pe_0 [&amp;_thead_th]:pt_0 [&amp;_thead_th]:c_accent [&amp;_thead_th]:white-space_nowrap [&amp;_thead_th]:bd-b_1px_dashed_token(colors.line-strong) [&amp;_tbody_th]:fw_normal [&amp;_tbody_tr:last-child_&gt;_*]:pb_0"><table><thead><tr><th scope="col"><strong>时间</strong></th><th scope="col"><strong>事件</strong></th><th scope="col"><strong>概要与影响范围</strong></th></tr></thead><tbody><tr><td><strong>2021年2月</strong></td><td><strong>“依赖混淆”漏洞披露</strong></td><td>安全研究员 Alex Birsan 利用<strong>依赖混淆</strong>（Dependency Confusion）技术，在 NPM/PyPI 上传与多家企业内部库同名的软件包，成功入侵了包括苹果、微软等35家大厂内部服务器 (<a href="https://www.sonatype.com/blog/pypi-flooded-with-over-1200-dependency-confusion-packages#:~:text=Dependency%20confusion%3A%20Year%20in%20review" rel="nofollow noopener noreferrer" target="_blank">PyPI flooded with 1,275 dependency confusion packages</a>)。这一演示引发业内对供应链风险的高度关注。</td></tr><tr><td><strong>2021年10月</strong></td><td><strong>UAParser.js 库遭劫持</strong></td><td>NPM上每周下载量超700万的流行库 <em>ua-parser-js</em> 被攻击者通过维护者账户入侵发布恶意版本 (<a href="https://www.sonatype.com/resources/vulnerability-timeline#:~:text=%23%23%20%20Popular%20%22ua,Attacked" rel="nofollow noopener noreferrer" target="_blank">A Timeline of SSC Attacks, Curated by Sonatype</a>)。受感染版本在安装时植入<strong>密码窃取木马</strong>和<strong>加密货币挖矿程序</strong>，波及大量开发者系统。</td></tr><tr><td><strong>2021年10月</strong></td><td><strong>假冒 Roblox 库投毒</strong></td><td>攻击者在 NPM 上传多个假冒 Roblox API 的软件包（如 <em>noblox.js-proxy</em>），内含混淆的恶意代码，安装后会植入<strong>木马和勒索软件</strong>等Payload (<a href="https://www.sonatype.com/resources/vulnerability-timeline#:~:text=,and%20has%20a%20Spooky%20Surprise" rel="nofollow noopener noreferrer" target="_blank">A Timeline of SSC Attacks, Curated by Sonatype</a>)。这些包下载数千次，显示出攻击者通过<strong>typosquatting</strong>手法诱骗游戏开发者。</td></tr><tr><td><strong>2021年11月</strong></td><td><strong>COA 与 RC 库连续劫持</strong></td><td>NPM上热门库 <em>coa</em>（每周下载数百万）和 <em>rc</em>（每周1400万下载）相继被入侵发布恶意版本。受害版本执行与 UAParser.js 案例类似的<strong>凭证窃取木马</strong>，一度导致全球众多使用 React 等框架的项目构建管线中断 (<a href="https://www.sonatype.com/resources/vulnerability-timeline#:~:text=,js" rel="nofollow noopener noreferrer" target="_blank">A Timeline of SSC Attacks, Curated by Sonatype</a>) (<a href="https://www.sonatype.com/resources/vulnerability-timeline#:~:text=,Is%20Hijacked%2C%20Too" rel="nofollow noopener noreferrer" target="_blank">A Timeline of SSC Attacks, Curated by Sonatype</a>)。官方调查认定原因均为维护者账户被盗用。</td></tr><tr><td><strong>2022年1月</strong></td><td><strong>Colors/Faker 开源库“自杀”</strong></td><td>著名的颜色格式库 <em>colors.js</em> 和测试数据生成库 <em>faker.js</em> 的作者出于抗议，在最新版本中注入无限循环等破坏性代码，导致包括Meta（Facebook）和亚马逊等公司在内的数千项目崩溃 (<a href="https://www.sonatype.com/resources/vulnerability-timeline#:~:text=Thousands%20of%20open%20source%20projects,companies%20exploiting%20open%20source" rel="nofollow noopener noreferrer" target="_blank">A Timeline of SSC Attacks, Curated by Sonatype</a>)（虽非外部攻击，但属于供应链投毒范畴）。</td></tr><tr><td><strong>2022年1月</strong></td><td><strong>PyPI 1,275个恶意包集中投放</strong></td><td>一名用户在1月23日一天内疯狂向 PyPI 发布了 <strong>1,275 个恶意软件包</strong> (<a href="https://www.sonatype.com/resources/vulnerability-timeline#:~:text=,Than%201%2C200%20Dependency%20Confusion%20Packages" rel="nofollow noopener noreferrer" target="_blank">A Timeline of SSC Attacks, Curated by Sonatype</a>)。这些包大多冒用知名项目或公司的名字（如 <em>xcryptography</em>、<em>Sagepay</em> 等），安装后收集主机名、IP等指纹信息并通过 DNS/HTTP 回传给攻击者 (<a href="https://www.sonatype.com/blog/pypi-flooded-with-over-1200-dependency-confusion-packages#:~:text=The%20,of%20these%20components%20are%20installed" rel="nofollow noopener noreferrer" target="_blank">PyPI flooded with 1,275 dependency confusion packages</a>) (<a href="https://www.sonatype.com/blog/pypi-flooded-with-over-1200-dependency-confusion-packages#:~:text=For%20DNS%3A%20.sub.deliverycontent,online" rel="nofollow noopener noreferrer" target="_blank">PyPI flooded with 1,275 dependency confusion packages</a>)。PyPI 管理员在收到报告后一小时内即下架了所有相关包 (<a href="https://www.sonatype.com/blog/pypi-flooded-with-over-1200-dependency-confusion-packages#:~:text=All%20of%20the%201%2C275%20were,an%20hour%20of%20our%20report" rel="nofollow noopener noreferrer" target="_blank">PyPI flooded with 1,275 dependency confusion packages</a>)。</td></tr><tr><td><strong>2022年3月</strong></td><td><strong>Node-ipc “抗议软件”事件</strong></td><td>前端构建常用库 <em>node-ipc</em> 的作者在 v10.1.1–10.1.3 版本中加入恶意代码：检测到客户端 IP 属于俄罗斯或白俄罗斯时，就<strong>擦除文件系统</strong>、用爱心表情覆盖文件 (<a href="https://www.zdnet.com/article/corrupted-open-source-software-enters-the-russian-battlefield/#:~:text=To%20be%20exact%2C%20Miller%20added,annoying%20to%20a%20system%20destroyer" rel="nofollow noopener noreferrer" target="_blank">Corrupted open-source software enters the Russian battlefield | ZDNET</a>) (<a href="https://www.zdnet.com/article/corrupted-open-source-software-enters-the-russian-battlefield/#:~:text=According%20to%20developer%20security%20company,8%2C%20critical" rel="nofollow noopener noreferrer" target="_blank">Corrupted open-source software enters the Russian battlefield | ZDNET</a>)。该库被 Vue CLI 等广泛依赖，导致大量用户系统遭破坏，并被赋予 CVE-2022-23812（CVSS 9.8） (<a href="https://www.zdnet.com/article/corrupted-open-source-software-enters-the-russian-battlefield/#:~:text=According%20to%20developer%20security%20company,8%2C%20critical" rel="nofollow noopener noreferrer" target="_blank">Corrupted open-source software enters the Russian battlefield | ZDNET</a>)。</td></tr><tr><td><strong>2022年10月</strong></td><td><strong>LofyGang 大规模投毒活动</strong></td><td>安全公司发现一个名为“LofyGang”的团伙在 NPM 上分发了将近 <strong>200 个恶意包</strong> (<a href="https://thehackernews.com/2022/10/lofygang-distributed-200-malicious-npm.html#:~:text=Multiple%20campaigns%20that%20distributed%20trojanized,single%20threat%20actor%20dubbed%20LofyGang" rel="nofollow noopener noreferrer" target="_blank">LofyGang Distributed ~200 Malicious NPM Packages to Steal Credit Card Data</a>)。这些包通过<strong>typosquatting</strong>和伪装常用库名称植入<strong>木马</strong>，窃取开发者的信用卡信息、Discord 账户以及游戏服务登录凭据，累计安装次数达数千次 (<a href="https://thehackernews.com/2022/10/lofygang-distributed-200-malicious-npm.html#:~:text=Multiple%20campaigns%20that%20distributed%20trojanized,single%20threat%20actor%20dubbed%20LofyGang" rel="nofollow noopener noreferrer" target="_blank">LofyGang Distributed ~200 Malicious NPM Packages to Steal Credit Card Data</a>)。这是一起持续一年多的有组织网络犯罪活动。</td></tr><tr><td><strong>2022年12月</strong></td><td><strong>PyTorch-nightly 依赖链攻击</strong></td><td>知名深度学习框架 PyTorch 披露其夜间版在 12月25–30日间遭遇<strong>依赖混淆式</strong>供应链攻击 (<a href="https://www.wiz.io/blog/malicious-pytorch-dependency-torchtriton-on-pypi-everything-you-need-to-know#:~:text=means%20that%20anyone%20who%20downloaded,and%20rotate%20any%20discovered%20keys" rel="nofollow noopener noreferrer" target="_blank">Malicious PyTorch dependency &#x27;torchtriton&#x27; on PyPI | Wiz Blog</a>)：攻击者在 PyPI 上注册了名为 <em>torchtriton</em> 的恶意包，与 PyTorch 夜ly 版所需的私有依赖同名，导致数千名通过 pip 安装 nightly 版的用户中招 (<a href="https://www.wiz.io/blog/malicious-pytorch-dependency-torchtriton-on-pypi-everything-you-need-to-know#:~:text=means%20that%20anyone%20who%20downloaded,and%20rotate%20any%20discovered%20keys" rel="nofollow noopener noreferrer" target="_blank">Malicious PyTorch dependency &#x27;torchtriton&#x27; on PyPI | Wiz Blog</a>)。恶意 <em>torchtriton</em> 包运行后收集系统上的环境变量和秘钥并上传至攻击者服务器，危及用户的云凭证安全。PyTorch 官方紧急发布警告并替换了该命名空间 (<a href="https://www.wiz.io/blog/malicious-pytorch-dependency-torchtriton-on-pypi-everything-you-need-to-know#:~:text=The%20creator%20of%20the%20copied,were%20stored%20on%20impacted%20resources" rel="nofollow noopener noreferrer" target="_blank">Malicious PyTorch dependency &#x27;torchtriton&#x27; on PyPI | Wiz Blog</a>)。</td></tr><tr><td><strong>2023年3月</strong></td><td><strong>“W4SP Stealer” 木马泛滥 PyPI</strong></td><td>安全研究员陆续发现 PyPI 上出现大量携带 <strong>W4SP Stealer</strong> 信息窃取木马的恶意包 (<a href="https://thehackernews.com/2022/12/w4sp-stealer-discovered-in-multiple.html#:~:text=Threat%20actors%20have%20published%20yet,malware%20on%20compromised%20developer%20machines" rel="nofollow noopener noreferrer" target="_blank">W4SP Stealer Discovered in Multiple PyPI Packages Under Various Names</a>)。这些木马别名众多（如 ANGEL Stealer、PURE Stealer 等），但本质均为 W4SP 家族，专门窃取用户密码、加密货币钱包和 Discord 令牌等信息 (<a href="https://thehackernews.com/2022/12/w4sp-stealer-discovered-in-multiple.html#:~:text=Interestingly%2C%20while%20the%20malware%20goes,be%20copies%20of%20W4SP%20Stealer" rel="nofollow noopener noreferrer" target="_blank">W4SP Stealer Discovered in Multiple PyPI Packages Under Various Names</a>)。一次报告就揭示了16个此类恶意包（如 <em>modulesecurity</em>、<em>easycordey</em> 等） (<a href="https://thehackernews.com/2022/12/w4sp-stealer-discovered-in-multiple.html#:~:text=The%2016%20rogue%20modules%20are,nowsys%2C%20upamonkws%2C%20captchaboy%2C%20and%20proxybooster" rel="nofollow noopener noreferrer" target="_blank">W4SP Stealer Discovered in Multiple PyPI Packages Under Various Names</a>)。PyPI 针对此类木马展开清理，并加强了上传检测。</td></tr><tr><td><strong>2023年8月</strong></td><td><strong>Lazarus 组织攻击 PyPI</strong></td><td>ReversingLabs 报告称朝鲜黑客组织 Lazarus 的分支在 PyPI 发布了逾两打（24个以上）伪装热门库的恶意包（代号“VMConnect”行动） (<a href="https://www.reversinglabs.com/blog/a-partial-history-of-software-supply-chain-attacks#:~:text=" rel="nofollow noopener noreferrer" target="_blank">Software Supply Chain Attacks: A (partial) History</a>)。这些包企图针对特定行业（如金融）用户，植入远程访问木马。据称该攻击与此前针对 NuGet 的类似活动相关联，显示出国家级黑客对开源供应链的兴趣。</td></tr><tr><td><strong>2024年及以后</strong></td><td><strong>持续的供应链威胁</strong></td><td>2024年以来，NPM 与 PyPI 上仍不断爆出新的投毒事件。例如2024年初发现假冒VS Code相关NPM包内含远控间谍软件 (<a href="https://www.sonatype.com/resources/vulnerability-timeline#:~:text=,altered%20ScreenConnect%20utility%20as%20spyware" rel="nofollow noopener noreferrer" target="_blank">A Timeline of SSC Attacks, Curated by Sonatype</a>)、假冒Solana库窃取加密钱包密钥的PyPI包 (<a href="https://www.sonatype.com/resources/vulnerability-timeline#:~:text=%23%23%20%20Ideal%20typosquat%20%27solana,steals%20your%20crypto%20wallet%20keys" rel="nofollow noopener noreferrer" target="_blank">A Timeline of SSC Attacks, Curated by Sonatype</a>)等。这表明供应链攻击已成常态化威胁，需要生态系统持续提高警惕和防御能力。</td></tr></tbody></table></div></div></figure>
<p>发 Twitter 吐槽了一下，结果吐槽的时候就看到一个推友遇到了一起供应链攻击事件。</p>
<p><a href="https://x.com/tcdwww/status/1914202659210359108" rel="nofollow noopener noreferrer" target="_blank"><img src="https://static.miantiao.me/share/ZUy0MY/twitter.jpeg" alt="Twitter"/></a></p>
<p>所幸 <a href="https://x.com/TBXark" rel="nofollow noopener noreferrer" target="_blank">@TBXark</a> 推荐了他的 <strong>MCP Proxy</strong> 项目，可以很方便的将 MCP Server 运行在 Docker 中。他最初的目的是把 MCP Server 运行在服务器上，减少客户端压力和方便移动端调用。 然而由于 Docker 天然的隔离特性，与我期望有沙箱的诉求不谋而合。</p>
<p>MCP Proxy 会在 Docker 中运行 MCP Servers 并转换为 MCP SSE 的协议，MCP 客户端的调用链路如下：</p>
<figure class="pos_relative min-w_0 bd_1px_dashed_token(colors.line-strong) p_1lh_2ch [&amp;_&gt;_figcaption]:pos_relative [&amp;_&gt;_figcaption]:w_fit-content [&amp;_&gt;_figcaption]:max-w_100% [&amp;_&gt;_figcaption]:m_-1.5lh_auto_0.5lh [&amp;_&gt;_figcaption]:px_1ch [&amp;_&gt;_figcaption]:bg_bg [&amp;_&gt;_figcaption]:c_accent [&amp;_&gt;_figcaption]:ta_center [&amp;_&gt;_figcaption]:ov-wrap_anywhere [&amp;_&gt;_div_&gt;_:first-child]:mt_0 [&amp;_&gt;_div_&gt;_:last-child]:mb_0"><figcaption><span aria-hidden="true">[ </span>通过 MCP Proxy 调用容器中的服务<span aria-hidden="true"> ]</span></figcaption><div><div class="d_grid gap_1lh"><p class="d_flex flex-wrap_wrap ai_baseline gap_0.5lh_1ch [&amp;_strong]:c_accent"><span class="bd_1px_solid_token(colors.line) px_1ch min-w_0 ov-wrap_anywhere">MCP 客户端</span><span> → </span><span class="bd_1px_solid_token(colors.line) px_1ch min-w_0 ov-wrap_anywhere"><strong>MCP Proxy（SSE）</strong></span><span> → </span><span class="bd_1px_solid_token(colors.line) px_1ch min-w_0 ov-wrap_anywhere">Docker 中的 MCP Server</span></p></div></div></figure>
<p>这样可以大大减小 npx 和 uvx 直接运行带来的任意文件读取风险。<em>如果部署在境外服务器, 还可以顺带解决网络的问题</em>。</p>
<p>但是当前还是可以读取到 <code>/config/config.json</code> 这个 MCP Proxy 的配置文件, 风险可控。同时也给开发者提了需求， config 文件配置 400 权限， npx 和 uvx 命令使用 nobody 用户运行。如果可以实现，将完美解决任意文件读取的问题。</p>
<h2>运行 MCP Proxy</h2>
<p><a href="https://github.com/TBXark/mcp-proxy" rel="nofollow noopener noreferrer" target="_blank"><img loading="lazy" decoding="async" src="https://github.html.zone/TBXark/mcp-proxy" alt="MCP Proxy"/></a></p>
<p>如果你自己有 VPS 部署了 Docker, 可以使用下面的命令运行 MCP Proxy。</p>
<pre class="kirari"><code class="language-bash"><span class="kirari-variable">docker</span> <span class="kirari-variable">run</span> -<span class="kirari-variable">d</span> -<span class="kirari-variable">p</span> <span class="kirari-number">9090</span>:<span class="kirari-number">9090</span> -<span class="kirari-variable">v</span> /<span class="kirari-variable">path</span>/<span class="kirari-variable">to</span>/<span class="kirari-variable">config</span>.<span class="kirari-variable">json</span>:/<span class="kirari-variable">config</span>/<span class="kirari-variable">config</span>.<span class="kirari-variable">json</span> <span class="kirari-variable">ghcr</span>.<span class="kirari-variable">io</span>/<span class="kirari-variable">tbxark</span>/<span class="kirari-variable">mcp-proxy</span>:<span class="kirari-variable">latest</span>
</code></pre>
<p>如果你没有自己的 VPS, 可以使用 <a href="https://404.li/claw" rel="nofollow noopener noreferrer" target="_blank"><strong>claw.cloud</strong></a> 提供的免费容器服务（每个月 $5 额度, GitHub 注册需满 180 天）。</p>
<p>由于 Claw 有容器大小的限制，我们需要使用下面的环境变量，配置 npx 和 uvx 的缓存目录，防止容器崩溃。</p>
<pre class="kirari"><code class="language-txt">UV_CACHE_DIR=/cache/uv
npm_config_cache=/cache/npm
</code></pre>
<p>同时在 <code>/cache</code> 路径下挂载 10G 的存储。 配置参考我的配置： 0.5c CPU, 512M 内存, 10G 硬盘。</p>
<p>最终的配置如下：</p>
<p><button type="button" data-zoomable="" aria-label="放大图片：Claw"><img loading="lazy" decoding="async" src="https://static.miantiao.me/share/g4KUgP/claw.jpg" alt="Claw"/></button></p>
<h2>配置 MCP Proxy</h2>
<p>MCP Proxy 的配置文件需要挂载在 <code>/config/config.json</code> 路径下，完整配置请参考 <a href="https://github.com/TBXark/mcp-proxy?tab=readme-ov-file#configurationonfiguration" rel="nofollow noopener noreferrer" target="_blank">https://github.com/TBXark/mcp-proxy?tab=readme-ov-file#configurationonfiguration</a>。</p>
<p>以下是我的配置，可以参考。</p>
<pre class="kirari"><code class="language-json"><span class="kirari-punctuation">{</span>
    <span class="kirari-property">&quot;mcpProxy&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-punctuation">{</span>
        <span class="kirari-property">&quot;baseURL&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&quot;https://mcp.miantiao.me&quot;</span><span class="kirari-punctuation">,</span>
        <span class="kirari-property">&quot;addr&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&quot;:9090&quot;</span><span class="kirari-punctuation">,</span>
        <span class="kirari-property">&quot;name&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&quot;MCP Proxy&quot;</span><span class="kirari-punctuation">,</span>
        <span class="kirari-property">&quot;version&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&quot;1.0.0&quot;</span><span class="kirari-punctuation">,</span>
        <span class="kirari-property">&quot;options&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-punctuation">{</span>
          <span class="kirari-property">&quot;panicIfInvalid&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-boolean">false</span><span class="kirari-punctuation">,</span>
          <span class="kirari-property">&quot;logEnabled&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-boolean">true</span><span class="kirari-punctuation">,</span>
          <span class="kirari-property">&quot;authTokens&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-punctuation">[</span>
            <span class="kirari-string">&quot;miantiao.me&quot;</span>
          <span class="kirari-punctuation">]</span>
        <span class="kirari-punctuation">}</span>
    <span class="kirari-punctuation">}</span><span class="kirari-punctuation">,</span>
    <span class="kirari-property">&quot;mcpServers&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-punctuation">{</span>
        <span class="kirari-property">&quot;github&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-punctuation">{</span>
            <span class="kirari-property">&quot;command&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&quot;npx&quot;</span><span class="kirari-punctuation">,</span>
            <span class="kirari-property">&quot;args&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-punctuation">[</span>
                <span class="kirari-string">&quot;-y&quot;</span><span class="kirari-punctuation">,</span>
                <span class="kirari-string">&quot;@modelcontextprotocol/server-github&quot;</span>
            <span class="kirari-punctuation">]</span><span class="kirari-punctuation">,</span>
            <span class="kirari-property">&quot;env&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-punctuation">{</span>
                <span class="kirari-property">&quot;GITHUB_PERSONAL_ACCESS_TOKEN&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&quot;&lt;YOUR_TOKEN&gt;&quot;</span>
            <span class="kirari-punctuation">}</span>
        <span class="kirari-punctuation">}</span><span class="kirari-punctuation">,</span>
        <span class="kirari-property">&quot;fetch&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-punctuation">{</span>
            <span class="kirari-property">&quot;command&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&quot;uvx&quot;</span><span class="kirari-punctuation">,</span>
            <span class="kirari-property">&quot;args&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-punctuation">[</span>
                <span class="kirari-string">&quot;mcp-server-fetch&quot;</span>
            <span class="kirari-punctuation">]</span>
        <span class="kirari-punctuation">}</span><span class="kirari-punctuation">,</span>
        <span class="kirari-property">&quot;amap&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-punctuation">{</span>
            <span class="kirari-property">&quot;url&quot;</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&quot;https://mcp.amap.com/sse?key=&lt;YOUR_TOKEN&gt;&quot;</span>
        <span class="kirari-punctuation">}</span>
    <span class="kirari-punctuation">}</span>
<span class="kirari-punctuation">}</span>
</code></pre>
<h2>调用 MCP proxy</h2>
<p>以 <a href="https://404.li/chatwise" rel="nofollow noopener noreferrer" target="_blank"><strong>ChatWise</strong></a> 调用 fetch 为例，直接配置 SSE 协议即可。</p>
<p><button type="button" data-zoomable="" aria-label="放大图片：fetch"><img loading="lazy" decoding="async" src="https://static.miantiao.me/share/mI3zIh/fetch.jpg" alt="fetch"/></button></p>
<p>是不是很简单，等 <a href="https://404.li/chatwise" rel="nofollow noopener noreferrer" target="_blank"><strong>ChatWise</strong></a> 出了移动端这样调用也是完全可用的。</p>
<p><button type="button" data-zoomable="" aria-label="放大图片：ChatWise"><img loading="lazy" decoding="async" src="https://static.miantiao.me/share/t43O9e/chatwise.jpg" alt="ChatWise"/></button></p><img src="https://c.statcounter.com/9823304/0/92a1d06c/1/" alt="" width="1" height="1" referrerPolicy="no-referrer-when-downgrade" />]]></content:encoded>
    </item>
    <item>
      <title>使用 Cloudflare Workers 合并音频文件</title>
      <link>https://miantiao.me/posts/cloudflare-audio-concat/</link>
      <guid>https://miantiao.me/posts/cloudflare-audio-concat/</guid>
      <description>使用 Cloudflare Workers 和 Browser Rendering 将多个音频文件拼接成一个音频文件</description>
      <pubDate>Sat, 19 Apr 2025 11:09:12 GMT</pubDate>
      <content:encoded><![CDATA[<p>最近把 <a href="https://hacker-news.agi.li/" rel="nofollow noopener noreferrer" target="_blank">Hacker News 中文播客</a> 改成了双人对话的形式，由于目前的语音合成模型还不能很好地处理双人对话，所以需要把每个人的音频文件拼接起来。</p>
<p>由于项目之前运行在 Cloudflare Workflow 的 Worker Runtime, 众所周知 Worker Runtime 缺少不少 Node.JS 特性，无法调用 C++ 扩展。而且 Cloudflare Container 还没有正式上线，所以只能使用 Browser Rendering 来实现。</p>
<p>合并音频文件一般都使用 FFMpeg 来做，现在 FFMpeg 也可以通过 WASM 在浏览器内运行了。所以大体的技术方案是：</p>
<figure class="pos_relative min-w_0 bd_1px_dashed_token(colors.line-strong) p_1lh_2ch [&amp;_&gt;_figcaption]:pos_relative [&amp;_&gt;_figcaption]:w_fit-content [&amp;_&gt;_figcaption]:max-w_100% [&amp;_&gt;_figcaption]:m_-1.5lh_auto_0.5lh [&amp;_&gt;_figcaption]:px_1ch [&amp;_&gt;_figcaption]:bg_bg [&amp;_&gt;_figcaption]:c_accent [&amp;_&gt;_figcaption]:ta_center [&amp;_&gt;_figcaption]:ov-wrap_anywhere [&amp;_&gt;_div_&gt;_:first-child]:mt_0 [&amp;_&gt;_div_&gt;_:last-child]:mb_0"><figcaption><span aria-hidden="true">[ </span>音频合并与存储链路<span aria-hidden="true"> ]</span></figcaption><div><div class="d_grid gap_1lh"><p class="d_flex flex-wrap_wrap ai_baseline gap_0.5lh_1ch [&amp;_strong]:c_accent"><span class="bd_1px_solid_token(colors.line) px_1ch min-w_0 ov-wrap_anywhere">Worker Binding</span><span> → </span><span class="bd_1px_solid_token(colors.line) px_1ch min-w_0 ov-wrap_anywhere">浏览器实例（打开音频合并页面，FFmpeg WASM）</span><span> → </span><span class="bd_1px_solid_token(colors.line) px_1ch min-w_0 ov-wrap_anywhere"><code>Blob</code></span><span> → </span><span class="bd_1px_solid_token(colors.line) px_1ch min-w_0 ov-wrap_anywhere">Worker</span><span> → </span><span class="bd_1px_solid_token(colors.line) px_1ch min-w_0 ov-wrap_anywhere">R2</span></p></div></div></figure>
<p>整体代码量不多，但是由于 Browser Rendering 只能远程调用，调试比较麻烦。</p>
<p>最终实现代码：</p>
<h3>浏览器内音频合并代码</h3>
<pre class="kirari"><code class="language-html"><span class="kirari-keyword">&lt;!doctype html&gt;</span>
<span class="kirari-punctuation">&lt;</span><span class="kirari-tag">html</span> <span class="kirari-attribute">lang</span><span class="kirari-operator">=</span><span class="kirari-string">&quot;en&quot;</span><span class="kirari-punctuation">&gt;</span>
  <span class="kirari-punctuation">&lt;</span><span class="kirari-tag">head</span><span class="kirari-punctuation">&gt;</span>
    <span class="kirari-punctuation">&lt;</span><span class="kirari-tag">meta</span> <span class="kirari-attribute">charset</span><span class="kirari-operator">=</span><span class="kirari-string">&quot;UTF-8&quot;</span> <span class="kirari-punctuation">/&gt;</span>
    <span class="kirari-punctuation">&lt;</span><span class="kirari-tag">meta</span> <span class="kirari-attribute">name</span><span class="kirari-operator">=</span><span class="kirari-string">&quot;viewport&quot;</span> <span class="kirari-attribute">content</span><span class="kirari-operator">=</span><span class="kirari-string">&quot;width=device-width, initial-scale=1.0&quot;</span> <span class="kirari-punctuation">/&gt;</span>
    <span class="kirari-punctuation">&lt;</span><span class="kirari-tag">title</span><span class="kirari-punctuation">&gt;</span>Audio<span class="kirari-punctuation">&lt;/</span><span class="kirari-tag">title</span><span class="kirari-punctuation">&gt;</span>
  <span class="kirari-punctuation">&lt;/</span><span class="kirari-tag">head</span><span class="kirari-punctuation">&gt;</span>
  <span class="kirari-punctuation">&lt;</span><span class="kirari-tag">body</span><span class="kirari-punctuation">&gt;</span>
    <span class="kirari-punctuation">&lt;</span><span class="kirari-tag">script</span><span class="kirari-punctuation">&gt;</span>
      <span class="kirari-keyword">const</span> <span class="kirari-function">concatAudioFilesOnBrowser</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">async</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">audioFiles</span><span class="kirari-punctuation">)</span> <span class="kirari-operator">=&gt;</span> <span class="kirari-punctuation">{</span>
        <span class="kirari-keyword">const</span> <span class="kirari-variable">script</span> <span class="kirari-operator">=</span> <span class="kirari-variable">document</span><span class="kirari-punctuation">.</span><span class="kirari-property">createElement</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;script&#x27;</span><span class="kirari-punctuation">)</span>
        <span class="kirari-variable">script</span><span class="kirari-punctuation">.</span><span class="kirari-property">src</span> <span class="kirari-operator">=</span> <span class="kirari-string">&#x27;https://unpkg.com/@ffmpeg/ffmpeg@0.11.6/dist/ffmpeg.min.js&#x27;</span>
        <span class="kirari-variable">document</span><span class="kirari-punctuation">.</span><span class="kirari-property">head</span><span class="kirari-punctuation">.</span><span class="kirari-property">appendChild</span><span class="kirari-punctuation">(</span><span class="kirari-variable">script</span><span class="kirari-punctuation">)</span>
        <span class="kirari-keyword">await</span> <span class="kirari-keyword">new</span> <span class="kirari-function">Promise</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">(</span><span class="kirari-variable">resolve</span><span class="kirari-punctuation">)</span> <span class="kirari-operator">=&gt;</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">script</span><span class="kirari-punctuation">.</span><span class="kirari-property">onload</span> <span class="kirari-operator">=</span> <span class="kirari-variable">resolve</span><span class="kirari-punctuation">)</span><span class="kirari-punctuation">)</span>

        <span class="kirari-keyword">const</span> <span class="kirari-punctuation">{</span> <span class="kirari-variable">createFFmpeg</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">fetchFile</span> <span class="kirari-punctuation">}</span> <span class="kirari-operator">=</span> <span class="kirari-variable">FFmpeg</span>
        <span class="kirari-keyword">const</span> <span class="kirari-variable">ffmpeg</span> <span class="kirari-operator">=</span> <span class="kirari-function">createFFmpeg</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">{</span> <span class="kirari-property">log</span><span class="kirari-punctuation">:</span> <span class="kirari-boolean">true</span> <span class="kirari-punctuation">}</span><span class="kirari-punctuation">)</span>

        <span class="kirari-keyword">await</span> <span class="kirari-variable">ffmpeg</span><span class="kirari-punctuation">.</span><span class="kirari-property">load</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span>

        <span class="kirari-comment">// Download and write each file to FFmpeg&#x27;s virtual file system</span>
        <span class="kirari-keyword">for</span> <span class="kirari-punctuation">(</span><span class="kirari-keyword">const</span> <span class="kirari-punctuation">[</span><span class="kirari-variable">index</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">audioFile</span><span class="kirari-punctuation">]</span> <span class="kirari-keyword">of</span> <span class="kirari-variable">audioFiles</span><span class="kirari-punctuation">.</span><span class="kirari-property">entries</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
          <span class="kirari-keyword">const</span> <span class="kirari-variable">audioData</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">await</span> <span class="kirari-function">fetchFile</span><span class="kirari-punctuation">(</span><span class="kirari-variable">audioFile</span><span class="kirari-punctuation">)</span>
          <span class="kirari-variable">ffmpeg</span><span class="kirari-punctuation">.</span><span class="kirari-property">FS</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;writeFile&#x27;</span><span class="kirari-punctuation">,</span> <span class="kirari-string">`input</span><span class="kirari-punctuation">${</span><span class="kirari-variable">index</span><span class="kirari-punctuation">}</span><span class="kirari-string">.mp3`</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">audioData</span><span class="kirari-punctuation">)</span>
        <span class="kirari-punctuation">}</span>

        <span class="kirari-comment">// Create a file list for ffmpeg concat</span>
        <span class="kirari-keyword">const</span> <span class="kirari-variable">fileList</span> <span class="kirari-operator">=</span> <span class="kirari-variable">audioFiles</span><span class="kirari-punctuation">.</span><span class="kirari-property">map</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">(</span><span class="kirari-variable">_</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">i</span><span class="kirari-punctuation">)</span> <span class="kirari-operator">=&gt;</span> <span class="kirari-string">`file &#x27;input</span><span class="kirari-punctuation">${</span><span class="kirari-variable">i</span><span class="kirari-punctuation">}</span><span class="kirari-string">.mp3&#x27;`</span><span class="kirari-punctuation">)</span><span class="kirari-punctuation">.</span><span class="kirari-property">join</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;\n&#x27;</span><span class="kirari-punctuation">)</span>
        <span class="kirari-variable">ffmpeg</span><span class="kirari-punctuation">.</span><span class="kirari-property">FS</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;writeFile&#x27;</span><span class="kirari-punctuation">,</span> <span class="kirari-string">&#x27;filelist.txt&#x27;</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">fileList</span><span class="kirari-punctuation">)</span>

        <span class="kirari-comment">// Execute FFmpeg command to concatenate files</span>
        <span class="kirari-keyword">await</span> <span class="kirari-variable">ffmpeg</span><span class="kirari-punctuation">.</span><span class="kirari-property">run</span><span class="kirari-punctuation">(</span>
          <span class="kirari-string">&#x27;-f&#x27;</span><span class="kirari-punctuation">,</span>
          <span class="kirari-string">&#x27;concat&#x27;</span><span class="kirari-punctuation">,</span>
          <span class="kirari-string">&#x27;-safe&#x27;</span><span class="kirari-punctuation">,</span>
          <span class="kirari-string">&#x27;0&#x27;</span><span class="kirari-punctuation">,</span>
          <span class="kirari-string">&#x27;-i&#x27;</span><span class="kirari-punctuation">,</span>
          <span class="kirari-string">&#x27;filelist.txt&#x27;</span><span class="kirari-punctuation">,</span>
          <span class="kirari-string">&#x27;-c:a&#x27;</span><span class="kirari-punctuation">,</span>
          <span class="kirari-string">&#x27;libmp3lame&#x27;</span><span class="kirari-punctuation">,</span>
          <span class="kirari-string">&#x27;-q:a&#x27;</span><span class="kirari-punctuation">,</span>
          <span class="kirari-string">&#x27;5&#x27;</span><span class="kirari-punctuation">,</span>
          <span class="kirari-string">&#x27;output.mp3&#x27;</span><span class="kirari-punctuation">,</span>
        <span class="kirari-punctuation">)</span>

        <span class="kirari-comment">// Read the output file</span>
        <span class="kirari-keyword">const</span> <span class="kirari-variable">data</span> <span class="kirari-operator">=</span> <span class="kirari-variable">ffmpeg</span><span class="kirari-punctuation">.</span><span class="kirari-property">FS</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;readFile&#x27;</span><span class="kirari-punctuation">,</span> <span class="kirari-string">&#x27;output.mp3&#x27;</span><span class="kirari-punctuation">)</span>

        <span class="kirari-comment">// Create a downloadable link</span>
        <span class="kirari-keyword">const</span> <span class="kirari-variable">blob</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">new</span> <span class="kirari-function">Blob</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">[</span><span class="kirari-variable">data</span><span class="kirari-punctuation">.</span><span class="kirari-property">buffer</span><span class="kirari-punctuation">]</span><span class="kirari-punctuation">,</span> <span class="kirari-punctuation">{</span> <span class="kirari-property">type</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&#x27;audio/mp3&#x27;</span> <span class="kirari-punctuation">}</span><span class="kirari-punctuation">)</span>

        <span class="kirari-comment">// Clean up</span>
        <span class="kirari-variable">audioFiles</span><span class="kirari-punctuation">.</span><span class="kirari-property">forEach</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">(</span><span class="kirari-variable">_</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">i</span><span class="kirari-punctuation">)</span> <span class="kirari-operator">=&gt;</span> <span class="kirari-punctuation">{</span>
          <span class="kirari-variable">ffmpeg</span><span class="kirari-punctuation">.</span><span class="kirari-property">FS</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;unlink&#x27;</span><span class="kirari-punctuation">,</span> <span class="kirari-string">`input</span><span class="kirari-punctuation">${</span><span class="kirari-variable">i</span><span class="kirari-punctuation">}</span><span class="kirari-string">.mp3`</span><span class="kirari-punctuation">)</span>
        <span class="kirari-punctuation">}</span><span class="kirari-punctuation">)</span>
        <span class="kirari-variable">ffmpeg</span><span class="kirari-punctuation">.</span><span class="kirari-property">FS</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;unlink&#x27;</span><span class="kirari-punctuation">,</span> <span class="kirari-string">&#x27;filelist.txt&#x27;</span><span class="kirari-punctuation">)</span>
        <span class="kirari-variable">ffmpeg</span><span class="kirari-punctuation">.</span><span class="kirari-property">FS</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;unlink&#x27;</span><span class="kirari-punctuation">,</span> <span class="kirari-string">&#x27;output.mp3&#x27;</span><span class="kirari-punctuation">)</span>

        <span class="kirari-keyword">return</span> <span class="kirari-variable">blob</span>
      <span class="kirari-punctuation">}</span>
    <span class="kirari-punctuation">&lt;/</span><span class="kirari-tag">script</span><span class="kirari-punctuation">&gt;</span>
  <span class="kirari-punctuation">&lt;/</span><span class="kirari-tag">body</span><span class="kirari-punctuation">&gt;</span>
<span class="kirari-punctuation">&lt;/</span><span class="kirari-tag">html</span><span class="kirari-punctuation">&gt;</span>
</code></pre>
<h3>Worker 调用代码</h3>
<pre class="kirari"><code class="language-ts"><span class="kirari-keyword">export</span> <span class="kirari-keyword">async</span> <span class="kirari-keyword">function</span> <span class="kirari-function">concatAudioFiles</span><span class="kirari-punctuation">(</span><span class="kirari-variable">audioFiles</span><span class="kirari-punctuation">:</span> <span class="kirari-type">string</span><span class="kirari-punctuation">[</span><span class="kirari-punctuation">]</span><span class="kirari-punctuation">,</span> <span class="kirari-property">BROWSER</span><span class="kirari-punctuation">:</span> <span class="kirari-variable">Fetcher</span><span class="kirari-punctuation">,</span> <span class="kirari-punctuation">{</span> <span class="kirari-variable">workerUrl</span> <span class="kirari-punctuation">}</span><span class="kirari-punctuation">:</span> <span class="kirari-punctuation">{</span> <span class="kirari-property">workerUrl</span><span class="kirari-punctuation">:</span> <span class="kirari-type">string</span> <span class="kirari-punctuation">}</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
  <span class="kirari-keyword">const</span> <span class="kirari-variable">browser</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">await</span> <span class="kirari-variable">puppeteer</span><span class="kirari-punctuation">.</span><span class="kirari-property">launch</span><span class="kirari-punctuation">(</span><span class="kirari-variable">BROWSER</span><span class="kirari-punctuation">)</span>
  <span class="kirari-keyword">const</span> <span class="kirari-variable">page</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">await</span> <span class="kirari-variable">browser</span><span class="kirari-punctuation">.</span><span class="kirari-property">newPage</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span>
  <span class="kirari-keyword">await</span> <span class="kirari-variable">page</span><span class="kirari-punctuation">.</span><span class="kirari-property">goto</span><span class="kirari-punctuation">(</span><span class="kirari-string">`</span><span class="kirari-punctuation">${</span><span class="kirari-variable">workerUrl</span><span class="kirari-punctuation">}</span><span class="kirari-string">/audio`</span><span class="kirari-punctuation">)</span>

  <span class="kirari-variable">console</span><span class="kirari-punctuation">.</span><span class="kirari-property">info</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;start concat audio files&#x27;</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">audioFiles</span><span class="kirari-punctuation">)</span>
  <span class="kirari-keyword">const</span> <span class="kirari-variable">fileUrl</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">await</span> <span class="kirari-variable">page</span><span class="kirari-punctuation">.</span><span class="kirari-property">evaluate</span><span class="kirari-punctuation">(</span><span class="kirari-keyword">async</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">audioFiles</span><span class="kirari-punctuation">)</span> <span class="kirari-operator">=&gt;</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-comment">// 此处 JS 运行在浏览器中</span>
    <span class="kirari-comment">// @ts-expect-error 浏览器内的对象</span>
    <span class="kirari-keyword">const</span> <span class="kirari-variable">blob</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">await</span> <span class="kirari-function">concatAudioFilesOnBrowser</span><span class="kirari-punctuation">(</span><span class="kirari-variable">audioFiles</span><span class="kirari-punctuation">)</span>

    <span class="kirari-keyword">const</span> <span class="kirari-variable">result</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">new</span> <span class="kirari-function">Promise</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">(</span><span class="kirari-variable">resolve</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">reject</span><span class="kirari-punctuation">)</span> <span class="kirari-operator">=&gt;</span> <span class="kirari-punctuation">{</span>
      <span class="kirari-keyword">const</span> <span class="kirari-variable">reader</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">new</span> <span class="kirari-function">FileReader</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span>
      <span class="kirari-variable">reader</span><span class="kirari-punctuation">.</span><span class="kirari-property">onloadend</span> <span class="kirari-operator">=</span> <span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span> <span class="kirari-operator">=&gt;</span> <span class="kirari-function">resolve</span><span class="kirari-punctuation">(</span><span class="kirari-variable">reader</span><span class="kirari-punctuation">.</span><span class="kirari-property">result</span><span class="kirari-punctuation">)</span>
      <span class="kirari-variable">reader</span><span class="kirari-punctuation">.</span><span class="kirari-property">onerror</span> <span class="kirari-operator">=</span> <span class="kirari-variable">reject</span>
      <span class="kirari-variable">reader</span><span class="kirari-punctuation">.</span><span class="kirari-property">readAsDataURL</span><span class="kirari-punctuation">(</span><span class="kirari-variable">blob</span><span class="kirari-punctuation">)</span>
    <span class="kirari-punctuation">}</span><span class="kirari-punctuation">)</span>
    <span class="kirari-keyword">return</span> <span class="kirari-keyword">await</span> <span class="kirari-variable">result</span>
  <span class="kirari-punctuation">}</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">audioFiles</span><span class="kirari-punctuation">)</span> <span class="kirari-keyword">as</span> <span class="kirari-type">string</span>

  <span class="kirari-variable">console</span><span class="kirari-punctuation">.</span><span class="kirari-property">info</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;concat audio files result&#x27;</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">fileUrl</span><span class="kirari-punctuation">.</span><span class="kirari-property">substring</span><span class="kirari-punctuation">(</span><span class="kirari-number">0</span><span class="kirari-punctuation">,</span> <span class="kirari-number">100</span><span class="kirari-punctuation">)</span><span class="kirari-punctuation">)</span>

  <span class="kirari-keyword">await</span> <span class="kirari-variable">browser</span><span class="kirari-punctuation">.</span><span class="kirari-property">close</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span>

  <span class="kirari-keyword">const</span> <span class="kirari-variable">response</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">await</span> <span class="kirari-function">fetch</span><span class="kirari-punctuation">(</span><span class="kirari-variable">fileUrl</span><span class="kirari-punctuation">)</span>
  <span class="kirari-keyword">return</span> <span class="kirari-keyword">await</span> <span class="kirari-variable">response</span><span class="kirari-punctuation">.</span><span class="kirari-property">blob</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span>
<span class="kirari-punctuation">}</span>

<span class="kirari-keyword">const</span> <span class="kirari-variable">audio</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">await</span> <span class="kirari-function">concatAudioFiles</span><span class="kirari-punctuation">(</span><span class="kirari-variable">audioFiles</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">env</span><span class="kirari-punctuation">.</span><span class="kirari-property">BROWSER</span><span class="kirari-punctuation">,</span> <span class="kirari-punctuation">{</span> <span class="kirari-property">workerUrl</span><span class="kirari-punctuation">:</span> <span class="kirari-variable">env</span><span class="kirari-punctuation">.</span><span class="kirari-property">HACKER_NEWS_WORKER_URL</span> <span class="kirari-punctuation">}</span><span class="kirari-punctuation">)</span>
<span class="kirari-keyword">return</span> <span class="kirari-keyword">new</span> <span class="kirari-function">Response</span><span class="kirari-punctuation">(</span><span class="kirari-variable">audio</span><span class="kirari-punctuation">)</span>
</code></pre>
<p>上面的代码基本是 Cursor 写的，最终效果可以去 <a href="https://github.com/miantiao-me/hacker-news/tree/main/worker" rel="nofollow noopener noreferrer" target="_blank">Hacker News 代码仓库</a> 看。</p><img src="https://c.statcounter.com/9823304/0/92a1d06c/1/" alt="" width="1" height="1" referrerPolicy="no-referrer-when-downgrade" />]]></content:encoded>
    </item>
    <item>
      <title>使用 AI 让博客排版更上一个台阶</title>
      <link>https://miantiao.me/posts/ai-blog-typography/</link>
      <guid>https://miantiao.me/posts/ai-blog-typography/</guid>
      <description>使用人工智能技术重新定义博客文章的视觉呈现，从简约到精致，从平凡到卓越。</description>
      <pubDate>Sun, 23 Mar 2025 13:39:55 GMT</pubDate>
      <content:encoded><![CDATA[<link rel="preload" as="image" href="https://static.miantiao.me/share/2025/wm70yi/hn-old.png"/><p>之前 Claude 3.7 发布，用它生成了一个邮箱 App 的设计图，效果还不错。</p>
<p>这周在 Twitter 上看到<a href="https://x.com/vista8/status/1901224129405338102" rel="nofollow noopener noreferrer" target="_blank">乔木老师分享</a>的生成网页、生成 PPT、生成 3D 教学动画、生成 SVG 海报的提示词，感觉很有意思。</p>
<p>这个周末把博客引擎升级了一下，使用 <strong>Astro + TailwindCSS + MDX</strong> 重构了博客。由于 MDX 有更强的扩展性，所以博客文章正文几乎可以展示任何内容。</p>
<p>于是使用乔木老师分享的提示词，修改了以后，用来生成博客文章的排版。效果出乎意料，从此不再为博客文章的排版而烦恼。</p>
<h2>排版前后对比</h2>
<h3><a href="/posts/hacker-news-podcast/">HackerNews Podcast</a></h3>
<p>修改前：</p>
<p><button type="button" data-zoomable="" aria-label="放大图片：HackerNews Podcast 文章排版修改前"><img src="https://static.miantiao.me/share/2025/wm70yi/hn-old.png" alt="HackerNews Podcast 文章排版修改前"/></button></p>
<p>修改后：</p>
<p><button type="button" data-zoomable="" aria-label="放大图片：HackerNews Podcast 文章排版修改后"><img loading="lazy" decoding="async" src="https://static.miantiao.me/share/2025/P9Rcfo/hn-new.png" alt="HackerNews Podcast 文章排版修改后"/></button></p>
<h3><a href="/posts/rss-beauty/">RSS Beauty</a></h3>
<p>修改前：</p>
<p><button type="button" data-zoomable="" aria-label="放大图片：RSS Beauty 文章排版修改前"><img loading="lazy" decoding="async" src="https://static.miantiao.me/share/2025/DWNmIh/rss-old.png" alt="RSS Beauty 文章排版修改前"/></button></p>
<p>修改后：</p>
<p><button type="button" data-zoomable="" aria-label="放大图片：RSS Beauty 文章排版修改后"><img loading="lazy" decoding="async" src="https://static.miantiao.me/share/2025/keIqtA/rss-new.png" alt="RSS Beauty 文章排版修改后"/></button></p>
<h3><a href="/posts/sink/">Sink</a></h3>
<p>修改前：</p>
<p><button type="button" data-zoomable="" aria-label="放大图片：Sink 文章排版修改前"><img loading="lazy" decoding="async" src="https://static.miantiao.me/share/2025/T2xVdS/sink-old.png" alt="Sink 文章排版修改前"/></button></p>
<p>修改后：</p>
<p><button type="button" data-zoomable="" aria-label="放大图片：Sink 文章排版修改后"><img loading="lazy" decoding="async" src="https://static.miantiao.me/share/2025/udluK7/sink-new.png" alt="Sink 文章排版修改后"/></button></p>
<h3><a href="/posts/looooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooong/">URL Longer</a></h3>
<p>修改前：</p>
<p><button type="button" data-zoomable="" aria-label="放大图片：URL Longer 文章排版修改前"><img loading="lazy" decoding="async" src="https://static.miantiao.me/share/2025/8KZoRB/long-old.png" alt="URL Longer 文章排版修改前"/></button></p>
<p>修改后：</p>
<p><button type="button" data-zoomable="" aria-label="放大图片：URL Longer 文章排版修改后"><img loading="lazy" decoding="async" src="https://static.miantiao.me/share/2025/rE7o6B/long-new.png" alt="URL Longer 文章排版修改后"/></button></p>
<h2>更多展示效果</h2>
<ul>
<li><a href="/posts/ai-remove-image-background/">浏览器本地使用 AI 移除图片背景</a></li>
<li><a href="/posts/broadcast-channel/">BroadcastChannel - 将你的 Telegram Channel 转为微博客</a></li>
<li><a href="/posts/dns-surf/">DNS.Surf - 检测域名在全球的 DNS 解析结果</a></li>
<li><a href="/posts/email-ml/">TempMail.Best - 最棒的一次性临时邮箱</a></li>
</ul>
<h2>提示词</h2>
<p>提示词源自<a href="https://x.com/vista8/status/1901224129405338102" rel="nofollow noopener noreferrer" target="_blank">乔木老师的分享</a>，在此基础上做了修改：</p>
<pre class="kirari"><code class="language-md"><span class="kirari-heading"># 生成文章网页</span>

你是一名专业的网页设计师和前端开发专家，对现代 Web 设计趋势和最佳实践有深入理解，尤其擅长创造具有极高审美价值的用户界面。你的设计作品不仅功能完备，而且在视觉上令人惊叹，能够给用户带来强烈的&quot;Aha-moment&quot;体验。

请根据最后提供的内容，设计一个<span class="kirari-punctuation">**</span>美观、现代、易读<span class="kirari-punctuation">**</span>的&quot;中文&quot;可视化网页。请充分发挥你的专业判断，选择最能体现内容精髓的设计风格、配色方案、排版和布局。

<span class="kirari-punctuation">**</span>设计目标：<span class="kirari-punctuation">**</span>

<span class="kirari-punctuation">*</span> <span class="kirari-punctuation">**</span>视觉吸引力：<span class="kirari-punctuation">**</span> 创造一个在视觉上令人印象深刻的网页，能够立即吸引用户的注意力，并激发他们的阅读兴趣。
<span class="kirari-punctuation">*</span> <span class="kirari-punctuation">**</span>可读性：<span class="kirari-punctuation">**</span> 确保内容清晰易读，无论在桌面端还是移动端，都能提供舒适的阅读体验。
<span class="kirari-punctuation">*</span> <span class="kirari-punctuation">**</span>信息传达：<span class="kirari-punctuation">**</span> 以一种既美观又高效的方式呈现信息，突出关键内容，引导用户理解核心思想。
<span class="kirari-punctuation">*</span> <span class="kirari-punctuation">**</span>情感共鸣:<span class="kirari-punctuation">**</span> 通过设计激发与内容主题相关的情感（例如，对于励志内容，激发积极向上的情绪；对于严肃内容，营造庄重、专业的氛围）。

<span class="kirari-punctuation">**</span>设计指导（请灵活运用，而非严格遵循）：<span class="kirari-punctuation">**</span>

<span class="kirari-punctuation">*</span> <span class="kirari-punctuation">**</span>整体风格：<span class="kirari-punctuation">**</span>
<span class="kirari-punctuation">  *</span> 可以考虑杂志风格、出版物风格，或者其他你认为合适的现代 Web 设计风格。
<span class="kirari-punctuation">  *</span> 配色参考 shadcn ui 的 Zinc 主题配色。
<span class="kirari-punctuation">  *</span> 目标是创造一个既有信息量，又有视觉吸引力的页面，就像一本精心设计的数字杂志或一篇深度报道。
<span class="kirari-punctuation">*</span> <span class="kirari-punctuation">**</span>Hero 模块（可选，但强烈建议）：<span class="kirari-punctuation">**</span>
<span class="kirari-punctuation">  *</span> 如果你认为合适，可以设计一个引人注目的 Hero 模块。
<span class="kirari-punctuation">  *</span> 它可以包含标题(h2)、副标题（p）、一段引人入胜的引言。
<span class="kirari-punctuation">*</span> <span class="kirari-punctuation">**</span>排版：<span class="kirari-punctuation">**</span>
<span class="kirari-punctuation">  *</span> 精心选择字体组合（衬线和无衬线），以提升中文阅读体验。
<span class="kirari-punctuation">  *</span> 利用不同的字号、字重、颜色和样式，创建清晰的视觉层次结构。
<span class="kirari-punctuation">  *</span> 可以考虑使用一些精致的排版细节（如首字下沉、悬挂标点）来提升整体质感。
<span class="kirari-punctuation">  *</span> Tabler icon 中有很多图标，选合适的点缀增加趣味性。 使用实例 <span class="kirari-string">`icon-[tabler--label]`</span>
<span class="kirari-punctuation">*</span> <span class="kirari-punctuation">**</span>配色方案：<span class="kirari-punctuation">**</span>
<span class="kirari-punctuation">  *</span> 选择一套既和谐又具有视觉冲击力的配色方案。
<span class="kirari-punctuation">  *</span> 考虑使用高对比度的颜色组合来突出重要元素。
<span class="kirari-punctuation">  *</span> 可以探索渐变、阴影等效果来增加视觉深度。
<span class="kirari-punctuation">*</span> <span class="kirari-punctuation">**</span>布局：<span class="kirari-punctuation">**</span>
<span class="kirari-punctuation">  *</span> 使用基于网格的布局系统来组织页面元素。
<span class="kirari-punctuation">  *</span> 充分利用负空间（留白），创造视觉平衡和呼吸感。
<span class="kirari-punctuation">  *</span> 可以考虑使用卡片、分割线、图标等视觉元素来分隔和组织内容。
<span class="kirari-punctuation">*</span> <span class="kirari-punctuation">**</span>调性：<span class="kirari-punctuation">**</span>整体风格精致, 营造一种高级感。

<span class="kirari-punctuation">**</span>技术规范：<span class="kirari-punctuation">**</span>

<span class="kirari-punctuation">*</span> 使用 tailwindCSS 定义样式。
<span class="kirari-punctuation">*</span> 不使用 JS , HTML 和 CSS 优先。
<span class="kirari-punctuation">*</span> 只生成正文区域，网页已经使用 <span class="kirari-string">`prose`</span> 类包裹了, 可以使用 <span class="kirari-string">`not-prose`</span> 突破限制。
<span class="kirari-punctuation">*</span> 实现完整的深色/浅色模式切换功能。
<span class="kirari-punctuation">*</span> 代码结构清晰、语义化，包含适当的注释。
<span class="kirari-punctuation">*</span> 实现完整的响应式，必须在所有设备上（手机、平板、桌面）完美展示。

<span class="kirari-punctuation">**</span>额外加分项：<span class="kirari-punctuation">**</span>

<span class="kirari-punctuation">*</span> <span class="kirari-punctuation">**</span>微交互：<span class="kirari-punctuation">**</span> 添加微妙而有意义的微交互效果来提升用户体验（例如，按钮悬停效果、卡片悬停效果、页面滚动效果）。
<span class="kirari-punctuation">*</span> <span class="kirari-punctuation">**</span>补充信息：<span class="kirari-punctuation">**</span> 可以主动搜索并补充其他重要信息或模块（例如，关键概念的解释、相关人物的介绍等），以增强用户对内容的理解。

<span class="kirari-punctuation">**</span>输出要求：<span class="kirari-punctuation">**</span>

<span class="kirari-punctuation">*</span> 输出一个独立的 MDX 文件，生成的语法符合 MDX 规范和 JSX 规范。
<span class="kirari-punctuation">*</span> 修改范围不要超出 mdx 文件。
<span class="kirari-punctuation">*</span> 不要在正文中出现标签，发布时间相关的信息。
<span class="kirari-punctuation">*</span> 外链增加 <span class="kirari-string">`nofollow`</span>, 在新窗口打开, 保留 <span class="kirari-string">`title`</span> 属性。
<span class="kirari-punctuation">*</span> 代码块不做任何修改，依旧使用 md 格式。
<span class="kirari-punctuation">*</span> 确保代码符合 W3C 标准，没有错误或警告。

请你像一个真正的设计师一样思考，充分发挥你的专业技能和创造力，打造一个令人惊艳的网页！

待处理内容：@miantiao<span class="kirari-punctuation">_</span>me
</code></pre><img src="https://c.statcounter.com/9823304/0/92a1d06c/1/" alt="" width="1" height="1" referrerPolicy="no-referrer-when-downgrade" />]]></content:encoded>
    </item>
    <item>
      <title>用播客的方式听 Hacker News</title>
      <link>https://miantiao.me/posts/hacker-news-podcast/</link>
      <guid>https://miantiao.me/posts/hacker-news-podcast/</guid>
      <description>利用AI和Cloudflare将Hacker News热门内容转换为中文播客，支持RSS订阅，方便随时随地收听科技资讯</description>
      <pubDate>Mon, 03 Mar 2025 12:40:37 GMT</pubDate>
      <content:encoded><![CDATA[<link rel="preload" as="image" href="https://github.html.zone/miantiao-me/hacker-news"/><h2>项目背景</h2>
<p><a href="https://news.ycombinator.com/" rel="nofollow noopener noreferrer" target="_blank">Hacker News</a> 是我一直关注的重要资讯来源，它能持续提供新奇有趣的极客资讯。之前我每天都会花半小时左右的时间来浏览。</p>
<p>过年期间，我注意到 DeepSeek 非常火，就尝试使用 Cloudflare Workflow 编写了一个工作流。但生成的内容都很短，而且接口也不稳定。后来尝试 GPT 4.0 系列模型，效果也不理想，于是就暂时搁置了。</p>
<p>Gemini 2.0 发布后，我尝试了一下，发现生成文章的效果还不错。于是我便开发了一个 Web 界面，并加入了 RSS 订阅功能，这样就可以在上班路上用泛用型播客 App 收听 Hacker News 的资讯了。</p>
<ul>
<li>在线预览：<a href="https://hacker-news.agi.li" rel="nofollow noopener noreferrer" target="_blank">hacker-news.agi.li</a></li>
<li>播客订阅：<a href="https://hacker-news.agi.li/rss.xml" rel="nofollow noopener noreferrer" target="_blank">hacker-news.agi.li/rss.xml</a></li>
<li>项目源码：<a href="https://github.com/miantiao-me/hacker-news" rel="nofollow noopener noreferrer" target="_blank">miantiao-me/hacker-news</a></li>
</ul>
<p><a href="https://github.com/miantiao-me/hacker-news" rel="nofollow noopener noreferrer" target="_blank"><img src="https://github.html.zone/miantiao-me/hacker-news" alt="miantiao-me/hacker-news 的 GitHub 仓库卡片"/></a></p>
<h2>主要特性</h2>
<ul>
<li>自动抓取 Hacker News 每日热门文章</li>
<li>使用 AI 智能总结文章内容和评论</li>
<li>通过 Edge TTS 生成中文播报</li>
<li>支持网页和播客 App 收听</li>
<li>每日自动更新</li>
<li>提供文章摘要和完整播报文本</li>
</ul>
<h2>技术栈</h2>
<figure class="pos_relative min-w_0 bd_1px_dashed_token(colors.line-strong) p_1lh_2ch [&amp;_&gt;_figcaption]:pos_relative [&amp;_&gt;_figcaption]:w_fit-content [&amp;_&gt;_figcaption]:max-w_100% [&amp;_&gt;_figcaption]:m_-1.5lh_auto_0.5lh [&amp;_&gt;_figcaption]:px_1ch [&amp;_&gt;_figcaption]:bg_bg [&amp;_&gt;_figcaption]:c_accent [&amp;_&gt;_figcaption]:ta_center [&amp;_&gt;_figcaption]:ov-wrap_anywhere [&amp;_&gt;_div_&gt;_:first-child]:mt_0 [&amp;_&gt;_div_&gt;_:last-child]:mb_0"><div><div role="region" aria-label="表格" tabindex="0" class="min-w_0 ov-x_auto focusVisible:ring_2px_solid_token(colors.accent) focusVisible:ring-o_0.5ch [&amp;_table]:d_table [&amp;_table]:m_0 [&amp;_table]:w_100% [&amp;_table]:bd-cl_separate [&amp;_table]:bd-sp_0 [&amp;_th,_&amp;_td]:min-w_10ch [&amp;_th,_&amp;_td]:p_0.5lh_1ch [&amp;_th,_&amp;_td]:bd_0 [&amp;_th,_&amp;_td]:bd-s_1px_dashed_token(colors.line) [&amp;_th,_&amp;_td]:fv-num_tabular-nums [&amp;_th,_&amp;_td]:ov-wrap_anywhere [&amp;_tr_&gt;_:first-child]:bd-s_0 [&amp;_tr_&gt;_:first-child]:ps_0 [&amp;_tr_&gt;_:last-child]:pe_0 [&amp;_thead_th]:pt_0 [&amp;_thead_th]:c_accent [&amp;_thead_th]:white-space_nowrap [&amp;_thead_th]:bd-b_1px_dashed_token(colors.line-strong) [&amp;_tbody_th]:fw_normal [&amp;_tbody_tr:last-child_&gt;_*]:pb_0"><table><thead><tr><th scope="col">用途</th><th scope="col">技术</th></tr></thead><tbody><tr><td>应用框架</td><td>Next.js</td></tr><tr><td>部署和运行环境</td><td>Cloudflare Workers</td></tr><tr><td>语音合成</td><td>Edge TTS</td></tr><tr><td>内容生成</td><td>OpenAI API</td></tr><tr><td>样式与组件</td><td>Tailwind CSS、shadcn/ui</td></tr></tbody></table></div></div></figure>
<h2>工作流程</h2>
<figure class="pos_relative min-w_0 bd_1px_dashed_token(colors.line-strong) p_1lh_2ch [&amp;_&gt;_figcaption]:pos_relative [&amp;_&gt;_figcaption]:w_fit-content [&amp;_&gt;_figcaption]:max-w_100% [&amp;_&gt;_figcaption]:m_-1.5lh_auto_0.5lh [&amp;_&gt;_figcaption]:px_1ch [&amp;_&gt;_figcaption]:bg_bg [&amp;_&gt;_figcaption]:c_accent [&amp;_&gt;_figcaption]:ta_center [&amp;_&gt;_figcaption]:ov-wrap_anywhere [&amp;_&gt;_div_&gt;_:first-child]:mt_0 [&amp;_&gt;_div_&gt;_:last-child]:mb_0"><figcaption><span aria-hidden="true">[ </span>从热门帖子到中文播客<span aria-hidden="true"> ]</span></figcaption><div><div class="[&amp;_&gt;_ol]:m_0 [&amp;_&gt;_ol]:pl_3ch [&amp;_&gt;_ol]:li-t_decimal-leading-zero [&amp;_&gt;_ol_&gt;_li]:pl_1ch [&amp;_&gt;_ol_&gt;_li::marker]:c_accent [&amp;_&gt;_ol_&gt;_li::marker]:fv-num_tabular-nums [&amp;_&gt;_ol_&gt;_li_+_li]:mt_1lh [&amp;_&gt;_ol_&gt;_li_&gt;_p_+_p]:mt_0.5lh"><ol>
<li>
<p>定时抓取</p>
<p>每日自动收集 Hacker News 上最受欢迎的帖子。</p>
</li>
<li>
<p>AI 生成文稿</p>
<p>通过 Gemini 2.0 将英文内容翻译并总结为中文摘要和播报文稿。</p>
</li>
<li>
<p>语音合成</p>
<p>通过 Edge TTS 将文稿转换为自然流畅的语音。</p>
</li>
<li>
<p>存储</p>
<p>将文本和音频存储到 Cloudflare R2 和 KV。</p>
</li>
<li>
<p>分发</p>
<p>通过 RSS Feed 和网页提供访问，可以在任何播客应用中订阅收听。</p>
</li>
</ol></div></div></figure>
<h2>未来计划</h2>
<p>目前 TTS 使用的是 Edge TTS，只有一个女声。理想情况下，使用男声和女声对话的形式可能会更好。豆包的 TTS 音色很不错，但它是收费的，等后续有时间，我会考虑改进这部分。</p>
<h2>推荐 Cloudflare Workflow</h2>
<p>最后推荐一下 <a href="https://developers.cloudflare.com/workflows/" rel="nofollow noopener noreferrer" target="_blank">Cloudflare Workflow</a>，一个很棒的工作流运行平台，高效且低成本。</p>
<p><button type="button" data-zoomable="" aria-label="放大图片：Cloudflare Workflow 控制台截图"><img loading="lazy" decoding="async" src="https://static.miantiao.me/share/2025/0ZROOg/postspark_export_2025-03-22_21-39-43.webp" alt="Cloudflare Workflow 控制台截图"/></button></p><img src="https://c.statcounter.com/9823304/0/92a1d06c/1/" alt="" width="1" height="1" referrerPolicy="no-referrer-when-downgrade" />]]></content:encoded>
    </item>
    <item>
      <title>RSS.Beauty - 让 RSS 变漂亮!</title>
      <link>https://miantiao.me/posts/rss-beauty/</link>
      <guid>https://miantiao.me/posts/rss-beauty/</guid>
      <description>基于XSLT技术的RSS美化工具，将普通RSS/Atom订阅源转换为美观易读的响应式界面，提升阅读体验</description>
      <pubDate>Tue, 31 Dec 2024 14:50:48 GMT</pubDate>
      <content:encoded><![CDATA[<link rel="preload" as="image" href="https://github.html.zone/miantiao-me/RSS.Beauty"/><blockquote>
<p>拖了快半年的工具终于做完了。</p>
</blockquote>
<p><a href="https://rss.beauty/" rel="nofollow noopener noreferrer" target="_blank">RSS.Beauty</a> 是一个基于 <strong>XSLT</strong> 技术的 RSS 美化工具，可以将普通的 RSS/Atom 订阅源转换成美观的阅读界面，让信息流变得更加优雅，阅读体验更加舒适。</p>
<h2>主要特性</h2>
<figure class="pos_relative min-w_0 bd_1px_dashed_token(colors.line-strong) p_1lh_2ch [&amp;_&gt;_figcaption]:pos_relative [&amp;_&gt;_figcaption]:w_fit-content [&amp;_&gt;_figcaption]:max-w_100% [&amp;_&gt;_figcaption]:m_-1.5lh_auto_0.5lh [&amp;_&gt;_figcaption]:px_1ch [&amp;_&gt;_figcaption]:bg_bg [&amp;_&gt;_figcaption]:c_accent [&amp;_&gt;_figcaption]:ta_center [&amp;_&gt;_figcaption]:ov-wrap_anywhere [&amp;_&gt;_div_&gt;_:first-child]:mt_0 [&amp;_&gt;_div_&gt;_:last-child]:mb_0"><div><div role="region" aria-label="表格" tabindex="0" class="min-w_0 ov-x_auto focusVisible:ring_2px_solid_token(colors.accent) focusVisible:ring-o_0.5ch [&amp;_table]:d_table [&amp;_table]:m_0 [&amp;_table]:w_100% [&amp;_table]:bd-cl_separate [&amp;_table]:bd-sp_0 [&amp;_th,_&amp;_td]:min-w_10ch [&amp;_th,_&amp;_td]:p_0.5lh_1ch [&amp;_th,_&amp;_td]:bd_0 [&amp;_th,_&amp;_td]:bd-s_1px_dashed_token(colors.line) [&amp;_th,_&amp;_td]:fv-num_tabular-nums [&amp;_th,_&amp;_td]:ov-wrap_anywhere [&amp;_tr_&gt;_:first-child]:bd-s_0 [&amp;_tr_&gt;_:first-child]:ps_0 [&amp;_tr_&gt;_:last-child]:pe_0 [&amp;_thead_th]:pt_0 [&amp;_thead_th]:c_accent [&amp;_thead_th]:white-space_nowrap [&amp;_thead_th]:bd-b_1px_dashed_token(colors.line-strong) [&amp;_tbody_th]:fw_normal [&amp;_tbody_tr:last-child_&gt;_*]:pb_0"><table><thead><tr><th scope="col">特性</th><th scope="col">说明</th></tr></thead><tbody><tr><td>精美的阅读界面</td><td>重新设计的视觉体验，让 RSS 阅读变得赏心悦目</td></tr><tr><td>支持 RSS 2.0 和 Atom 1.0</td><td>兼容主流订阅格式，无需担心格式问题</td></tr><tr><td>响应式设计</td><td>适配各种设备，移动端阅读同样舒适</td></tr><tr><td>一键订阅</td><td>快速添加到主流 RSS 阅读器</td></tr><tr><td>支持自部署</td><td>可部署到自己的服务器，保持完全的控制权</td></tr></tbody></table></div></div></figure>
<h2>快速开始</h2>
<p>访问 <a href="https://rss.beauty" rel="nofollow noopener noreferrer" target="_blank">RSS.Beauty</a> 并输入任意 RSS 订阅源链接即可体验。以<a href="https://feed.miantiao.me/" rel="nofollow noopener noreferrer" target="_blank">本站订阅源</a>为例，<a href="https://rss.beauty/rss?url=https%3A%2F%2Ffeed.miantiao.me%2F" rel="nofollow noopener noreferrer" target="_blank">查看美化效果</a>。</p>
<h2>技术栈</h2>
<ul>
<li><a href="https://astro.build" rel="nofollow noopener noreferrer" target="_blank">Astro</a></li>
<li><a href="https://tailwindcss.com" rel="nofollow noopener noreferrer" target="_blank">TailwindCSS</a></li>
<li><a href="https://www.w3.org/TR/xslt/" rel="nofollow noopener noreferrer" target="_blank">XSLT</a></li>
</ul>
<h2>开源</h2>
<p>RSS.Beauty 是一个开源项目，欢迎贡献代码或提出建议：<a href="https://github.com/miantiao-me/RSS.Beauty" rel="nofollow noopener noreferrer" target="_blank">miantiao-me/RSS.Beauty</a>。</p>
<p><a href="https://github.com/miantiao-me/RSS.Beauty" rel="nofollow noopener noreferrer" target="_blank"><img src="https://github.html.zone/miantiao-me/RSS.Beauty" alt="miantiao-me/RSS.Beauty 的 GitHub 仓库卡片"/></a></p><img src="https://c.statcounter.com/9823304/0/92a1d06c/1/" alt="" width="1" height="1" referrerPolicy="no-referrer-when-downgrade" />]]></content:encoded>
    </item>
    <item>
      <title>在浏览器中轻松运行 Python 程序</title>
      <link>https://miantiao.me/posts/run-python-in-browser/</link>
      <guid>https://miantiao.me/posts/run-python-in-browser/</guid>
      <description>利用WebAssembly和Pyodide技术在浏览器中运行Python程序，以Microsoft的MarkItDown工具为例，实现无需安装即可将Office文件转换为Markdown的在线服务。</description>
      <pubDate>Sat, 21 Dec 2024 09:57:34 GMT</pubDate>
      <content:encoded><![CDATA[<link rel="preload" as="image" href="https://www.html.zone/markitdown.png"/><p>最近，微软开源了一个名为 <a href="https://github.com/microsoft/markitdown" rel="nofollow noopener noreferrer" target="_blank">MarkItDown</a> 的程序，可以将 Office 文件转换为 Markdown 格式。这个项目一经发布就迅速登上了 GitHub 热门榜。</p>
<p>然而，由于 MarkItDown 是一个 Python 程序，对于非技术用户来说使用起来可能有些困难。为了解决这个问题，我想到了利用 WebAssembly 技术在浏览器中直接运行 Python 代码。</p>
<p>在浏览器内运行 Python 的开源程序是 Pyodide，使用 WebAssembly 移植了 CPython，所以 Python 的语法都是支持的。 Cloudflare 的 Python Worker 也使用的 Pyodide。</p>
<blockquote>
<p>Pyodide 是 CPython 的一个移植版本，用于 WebAssembly/Emscripten。</p>
<p>Pyodide 使得在浏览器中使用 micropip 安装和运行 Python 包成为可能。任何在 PyPi 上有可用 wheel 文件的纯 Python 包都被支持。</p>
<p>许多具有 C 扩展的包也已被移植以供 Pyodide 使用。这些包括许多通用包，如 regex、PyYAML、lxml，以及包括 NumPy、pandas、SciPy、Matplotlib 和 scikit-learn 在内的科学 Python 包。Pyodide 配备了强大的 JavaScript ⟺ Python 外部函数接口，使得您可以在代码中自由地混合这两种语言，几乎没有摩擦。这包括对错误处理、async/await 的全面支持，以及更多功能。</p>
<p>在浏览器中使用时，Python 可以完全访问 Web API。</p>
</blockquote>
<p>尝试了一下运行 MarkItDown 没想到异常的顺利，看来 WebAssembly 真的是浏览器的未来。</p>
<p>遇到的主要挑战和解决方案：</p>
<ol>
<li>
<p><strong>文件传输问题</strong>：如何将用户选择的文件传递给 Worker 中的 Python 运行时？</p>
<ul>
<li>解决方案：利用 Pyodide 提供的方案，将浏览器文件转换为 ArrayBuffer，然后写入 Emscripten 文件系统的本地缓存。</li>
</ul>
</li>
<li>
<p><strong>依赖安装问题</strong>：PyPI 在中国大陆访问受限。</p>
<ul>
<li>解决方案：使用 Cloudflare 搭建 PyPI 镜像，详见：<a href="https://github.com/miantiao-me/cloudflare-pypi-mirror" rel="nofollow noopener noreferrer" target="_blank">Cloudflare PyPI Mirror</a>。</li>
</ul>
</li>
</ol>
<figure class="pos_relative min-w_0 bd_1px_dashed_token(colors.line-strong) p_1lh_2ch [&amp;_&gt;_figcaption]:pos_relative [&amp;_&gt;_figcaption]:w_fit-content [&amp;_&gt;_figcaption]:max-w_100% [&amp;_&gt;_figcaption]:m_-1.5lh_auto_0.5lh [&amp;_&gt;_figcaption]:px_1ch [&amp;_&gt;_figcaption]:bg_bg [&amp;_&gt;_figcaption]:c_accent [&amp;_&gt;_figcaption]:ta_center [&amp;_&gt;_figcaption]:ov-wrap_anywhere [&amp;_&gt;_div_&gt;_:first-child]:mt_0 [&amp;_&gt;_div_&gt;_:last-child]:mb_0"><figcaption><span aria-hidden="true">[ </span>浏览器文件到 Markdown 的转换链路<span aria-hidden="true"> ]</span></figcaption><div><div class="d_grid gap_1lh"><p class="d_flex flex-wrap_wrap ai_baseline gap_0.5lh_1ch [&amp;_strong]:c_accent"><span class="bd_1px_solid_token(colors.line) px_1ch min-w_0 ov-wrap_anywhere">浏览器文件</span><span> → </span><span class="bd_1px_solid_token(colors.line) px_1ch min-w_0 ov-wrap_anywhere"><code>ArrayBuffer</code></span><span> → </span><span class="bd_1px_solid_token(colors.line) px_1ch min-w_0 ov-wrap_anywhere">Emscripten 文件系统</span><span> → </span><span class="bd_1px_solid_token(colors.line) px_1ch min-w_0 ov-wrap_anywhere"><strong>Pyodide / MarkItDown</strong></span><span> → </span><span class="bd_1px_solid_token(colors.line) px_1ch min-w_0 ov-wrap_anywhere">Markdown</span></p></div></div></figure>
<p>最终，我们成功实现了一个完全运行在浏览器中的 MarkItDown 工具。欢迎访问 <a href="https://www.html.zone/markitdown/" rel="nofollow noopener noreferrer" target="_blank">Office File to Markdown</a> 进行体验。</p>
<p><a href="https://www.html.zone/markitdown/" rel="nofollow noopener noreferrer" target="_blank"><img src="https://www.html.zone/markitdown.png" alt="Office File to Markdown"/></a></p>
<p>最后放出一下 Worker 中运行 Python 的核心代码：</p>
<pre class="kirari"><code class="language-js"><span class="kirari-function">importScripts</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;https://testingcf.jsdelivr.net/pyodide/v0.26.4/full/pyodide.js&#x27;</span><span class="kirari-punctuation">)</span>

<span class="kirari-comment">// npmmirror 支持 pyodide ，但是不支持 pyodide 下的 zip 包</span>
<span class="kirari-comment">// importScripts(&#x27;https://registry.npmmirror.com/pyodide/0.26.4/files/pyodide.js&#x27;)</span>

<span class="kirari-keyword">async</span> <span class="kirari-keyword">function</span> <span class="kirari-function">loadPyodideAndPackages</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
  <span class="kirari-keyword">const</span> <span class="kirari-variable">pyodide</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">await</span> <span class="kirari-function">loadPyodide</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span>
  <span class="kirari-variable">globalThis</span><span class="kirari-punctuation">.</span><span class="kirari-property">pyodide</span> <span class="kirari-operator">=</span> <span class="kirari-variable">pyodide</span>

  <span class="kirari-keyword">await</span> <span class="kirari-variable">pyodide</span><span class="kirari-punctuation">.</span><span class="kirari-property">loadPackage</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;micropip&#x27;</span><span class="kirari-punctuation">)</span>

  <span class="kirari-keyword">const</span> <span class="kirari-variable">micropip</span> <span class="kirari-operator">=</span> <span class="kirari-variable">pyodide</span><span class="kirari-punctuation">.</span><span class="kirari-property">pyimport</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;micropip&#x27;</span><span class="kirari-punctuation">)</span>

  <span class="kirari-comment">// 需要支持 PEP 691 和跨域， 目前 tuna 支持 PEP 691，但不支持跨域 https://github.com/tuna/issues/issues/2092</span>
  <span class="kirari-comment">// micropip.set_index_urls([</span>
  <span class="kirari-comment">//   &#x27;https://pypi.your.domains/pypi/simple&#x27;,</span>
  <span class="kirari-comment">// ])</span>

  <span class="kirari-keyword">await</span> <span class="kirari-variable">micropip</span><span class="kirari-punctuation">.</span><span class="kirari-property">install</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;markitdown==0.0.1a2&#x27;</span><span class="kirari-punctuation">)</span>
<span class="kirari-punctuation">}</span>

<span class="kirari-keyword">const</span> <span class="kirari-variable">pyodideReadyPromise</span> <span class="kirari-operator">=</span> <span class="kirari-function">loadPyodideAndPackages</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span>

<span class="kirari-variable">globalThis</span><span class="kirari-punctuation">.</span><span class="kirari-property">onmessage</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">async</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">event</span><span class="kirari-punctuation">)</span> <span class="kirari-operator">=&gt;</span> <span class="kirari-punctuation">{</span>
  <span class="kirari-keyword">await</span> <span class="kirari-variable">pyodideReadyPromise</span>

  <span class="kirari-keyword">const</span> <span class="kirari-variable">file</span> <span class="kirari-operator">=</span> <span class="kirari-variable">event</span><span class="kirari-punctuation">.</span><span class="kirari-property">data</span>
  <span class="kirari-keyword">try</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-variable">console</span><span class="kirari-punctuation">.</span><span class="kirari-property">log</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;file&#x27;</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">file</span><span class="kirari-punctuation">)</span>
    <span class="kirari-keyword">const</span> <span class="kirari-variable">startTime</span> <span class="kirari-operator">=</span> <span class="kirari-variable">Date</span><span class="kirari-punctuation">.</span><span class="kirari-property">now</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span>
    <span class="kirari-variable">globalThis</span><span class="kirari-punctuation">.</span><span class="kirari-property">pyodide</span><span class="kirari-punctuation">.</span><span class="kirari-property">FS</span><span class="kirari-punctuation">.</span><span class="kirari-property">writeFile</span><span class="kirari-punctuation">(</span><span class="kirari-string">`/</span><span class="kirari-punctuation">${</span><span class="kirari-variable">file</span><span class="kirari-punctuation">.</span><span class="kirari-property">filename</span><span class="kirari-punctuation">}</span><span class="kirari-string">`</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">file</span><span class="kirari-punctuation">.</span><span class="kirari-property">buffer</span><span class="kirari-punctuation">)</span>

    <span class="kirari-keyword">await</span> <span class="kirari-variable">globalThis</span><span class="kirari-punctuation">.</span><span class="kirari-property">pyodide</span><span class="kirari-punctuation">.</span><span class="kirari-property">runPythonAsync</span><span class="kirari-punctuation">(</span><span class="kirari-string">`
from markitdown import MarkItDown

markitdown = MarkItDown()

result = markitdown.convert(&quot;/</span><span class="kirari-punctuation">${</span><span class="kirari-variable">file</span><span class="kirari-punctuation">.</span><span class="kirari-property">filename</span><span class="kirari-punctuation">}</span><span class="kirari-string">&quot;)
print(result.text_content)

with open(&quot;/</span><span class="kirari-punctuation">${</span><span class="kirari-variable">file</span><span class="kirari-punctuation">.</span><span class="kirari-property">filename</span><span class="kirari-punctuation">}</span><span class="kirari-string">.md&quot;, &quot;w&quot;) as file:
  file.write(result.text_content)
`</span><span class="kirari-punctuation">)</span>
    <span class="kirari-variable">globalThis</span><span class="kirari-punctuation">.</span><span class="kirari-property">postMessage</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">{</span>
      <span class="kirari-property">filename</span><span class="kirari-punctuation">:</span> <span class="kirari-string">`</span><span class="kirari-punctuation">${</span><span class="kirari-variable">file</span><span class="kirari-punctuation">.</span><span class="kirari-property">filename</span><span class="kirari-punctuation">}</span><span class="kirari-string">.md`</span><span class="kirari-punctuation">,</span>
      <span class="kirari-property">content</span><span class="kirari-punctuation">:</span> <span class="kirari-variable">globalThis</span><span class="kirari-punctuation">.</span><span class="kirari-property">pyodide</span><span class="kirari-punctuation">.</span><span class="kirari-property">FS</span><span class="kirari-punctuation">.</span><span class="kirari-property">readFile</span><span class="kirari-punctuation">(</span><span class="kirari-string">`/</span><span class="kirari-punctuation">${</span><span class="kirari-variable">file</span><span class="kirari-punctuation">.</span><span class="kirari-property">filename</span><span class="kirari-punctuation">}</span><span class="kirari-string">.md`</span><span class="kirari-punctuation">,</span> <span class="kirari-punctuation">{</span> <span class="kirari-property">encoding</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&#x27;utf8&#x27;</span> <span class="kirari-punctuation">}</span><span class="kirari-punctuation">)</span><span class="kirari-punctuation">,</span>
      <span class="kirari-property">time</span><span class="kirari-punctuation">:</span> <span class="kirari-variable">Date</span><span class="kirari-punctuation">.</span><span class="kirari-property">now</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span> <span class="kirari-operator">-</span> <span class="kirari-variable">startTime</span><span class="kirari-punctuation">,</span>
    <span class="kirari-punctuation">}</span><span class="kirari-punctuation">)</span>
  <span class="kirari-punctuation">}</span>
  <span class="kirari-keyword">catch</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">error</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-variable">globalThis</span><span class="kirari-punctuation">.</span><span class="kirari-property">postMessage</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">{</span> <span class="kirari-property">error</span><span class="kirari-punctuation">:</span> <span class="kirari-variable">error</span><span class="kirari-punctuation">.</span><span class="kirari-property">message</span> <span class="kirari-operator">||</span> <span class="kirari-string">&#x27;convert error&#x27;</span><span class="kirari-punctuation">,</span> <span class="kirari-property">filename</span><span class="kirari-punctuation">:</span> <span class="kirari-variable">file</span><span class="kirari-punctuation">.</span><span class="kirari-property">filename</span> <span class="kirari-punctuation">}</span><span class="kirari-punctuation">)</span>
  <span class="kirari-punctuation">}</span>
<span class="kirari-punctuation">}</span>
</code></pre><img src="https://c.statcounter.com/9823304/0/92a1d06c/1/" alt="" width="1" height="1" referrerPolicy="no-referrer-when-downgrade" />]]></content:encoded>
    </item>
    <item>
      <title>使用 Cloudflare Snippets 搭建一个不限流量的 Docker 镜像</title>
      <link>https://miantiao.me/posts/cloudflare-docker-mirror/</link>
      <guid>https://miantiao.me/posts/cloudflare-docker-mirror/</guid>
      <description>利用Cloudflare Snippets搭建高效无限流量的Docker镜像加速服务，支持多种容器注册表，包括Docker Hub和GitHub Container Registry</description>
      <pubDate>Sat, 21 Dec 2024 08:17:40 GMT</pubDate>
      <content:encoded><![CDATA[<p>Cloudflare Workers 搭建 Docker 镜像个人使用请求数小没啥问题。但是如果公开使用，大量的请求数还是会产生费用。</p>
<p>其实 Cloudflare 还有一个更轻量的 JS Runtime: Cloudflare Snippets, 但是也有更严格的限制：CPU 执行时间 5 ms，最大内存 2M, 最大代码量 32K。 不过拿来重写请求足够了。</p>
<p>遗憾的是 Cloudflare Snippets 目前还未对 Free 计划开放，不过<a href="https://blog.cloudflare.com/zh-cn/snippets-announcement/" rel="nofollow noopener noreferrer" target="_blank">他们博客说 Free 计划可以建 5 个 Snippets</a>。</p>
<p>如果你有 Pro 计划，拿 Cloudflare Workers 的代码稍微修改一下就可以运行, 支持 Docker Hub, Google Container Registry, GitHub Container Registry, Amazon Elastic Container Registry, Kubernetes Container Registry, Quay, Cloudsmith。</p>
<p>修改后的代码：</p>
<pre class="kirari"><code class="language-js"><span class="kirari-comment">// 原代码: https://github.com/ciiiii/cloudflare-docker-proxy/blob/master/src/index.js</span>

<span class="kirari-keyword">const</span> <span class="kirari-variable">CUSTOM_DOMAIN</span> <span class="kirari-operator">=</span> <span class="kirari-string">&#x27;your.domains&#x27;</span>
<span class="kirari-keyword">const</span> <span class="kirari-variable">MODE</span> <span class="kirari-operator">=</span> <span class="kirari-string">&#x27;production&#x27;</span>

<span class="kirari-keyword">const</span> <span class="kirari-variable">dockerHub</span> <span class="kirari-operator">=</span> <span class="kirari-string">&#x27;https://registry-1.docker.io&#x27;</span>

<span class="kirari-keyword">const</span> <span class="kirari-variable">routes</span> <span class="kirari-operator">=</span> <span class="kirari-punctuation">{</span>
  <span class="kirari-comment">// production</span>
  <span class="kirari-punctuation">[</span><span class="kirari-string">`docker.</span><span class="kirari-punctuation">${</span><span class="kirari-variable">CUSTOM_DOMAIN</span><span class="kirari-punctuation">}</span><span class="kirari-string">`</span><span class="kirari-punctuation">]</span><span class="kirari-punctuation">:</span> <span class="kirari-variable">dockerHub</span><span class="kirari-punctuation">,</span>
  <span class="kirari-punctuation">[</span><span class="kirari-string">`quay.</span><span class="kirari-punctuation">${</span><span class="kirari-variable">CUSTOM_DOMAIN</span><span class="kirari-punctuation">}</span><span class="kirari-string">`</span><span class="kirari-punctuation">]</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&#x27;https://quay.io&#x27;</span><span class="kirari-punctuation">,</span>
  <span class="kirari-punctuation">[</span><span class="kirari-string">`gcr.</span><span class="kirari-punctuation">${</span><span class="kirari-variable">CUSTOM_DOMAIN</span><span class="kirari-punctuation">}</span><span class="kirari-string">`</span><span class="kirari-punctuation">]</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&#x27;https://gcr.io&#x27;</span><span class="kirari-punctuation">,</span>
  <span class="kirari-punctuation">[</span><span class="kirari-string">`k8s-gcr.</span><span class="kirari-punctuation">${</span><span class="kirari-variable">CUSTOM_DOMAIN</span><span class="kirari-punctuation">}</span><span class="kirari-string">`</span><span class="kirari-punctuation">]</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&#x27;https://k8s.gcr.io&#x27;</span><span class="kirari-punctuation">,</span>
  <span class="kirari-punctuation">[</span><span class="kirari-string">`k8s.</span><span class="kirari-punctuation">${</span><span class="kirari-variable">CUSTOM_DOMAIN</span><span class="kirari-punctuation">}</span><span class="kirari-string">`</span><span class="kirari-punctuation">]</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&#x27;https://registry.k8s.io&#x27;</span><span class="kirari-punctuation">,</span>
  <span class="kirari-punctuation">[</span><span class="kirari-string">`ghcr.</span><span class="kirari-punctuation">${</span><span class="kirari-variable">CUSTOM_DOMAIN</span><span class="kirari-punctuation">}</span><span class="kirari-string">`</span><span class="kirari-punctuation">]</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&#x27;https://ghcr.io&#x27;</span><span class="kirari-punctuation">,</span>
  <span class="kirari-punctuation">[</span><span class="kirari-string">`cloudsmith.</span><span class="kirari-punctuation">${</span><span class="kirari-variable">CUSTOM_DOMAIN</span><span class="kirari-punctuation">}</span><span class="kirari-string">`</span><span class="kirari-punctuation">]</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&#x27;https://docker.cloudsmith.io&#x27;</span><span class="kirari-punctuation">,</span>
  <span class="kirari-punctuation">[</span><span class="kirari-string">`ecr.</span><span class="kirari-punctuation">${</span><span class="kirari-variable">CUSTOM_DOMAIN</span><span class="kirari-punctuation">}</span><span class="kirari-string">`</span><span class="kirari-punctuation">]</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&#x27;https://public.ecr.aws&#x27;</span><span class="kirari-punctuation">,</span>

  <span class="kirari-comment">// staging</span>
  <span class="kirari-punctuation">[</span><span class="kirari-string">`docker-staging.</span><span class="kirari-punctuation">${</span><span class="kirari-variable">CUSTOM_DOMAIN</span><span class="kirari-punctuation">}</span><span class="kirari-string">`</span><span class="kirari-punctuation">]</span><span class="kirari-punctuation">:</span> <span class="kirari-variable">dockerHub</span><span class="kirari-punctuation">,</span>
<span class="kirari-punctuation">}</span>

<span class="kirari-keyword">async</span> <span class="kirari-keyword">function</span> <span class="kirari-function">handleRequest</span><span class="kirari-punctuation">(</span><span class="kirari-variable">request</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
  <span class="kirari-keyword">const</span> <span class="kirari-variable">url</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">new</span> <span class="kirari-function">URL</span><span class="kirari-punctuation">(</span><span class="kirari-variable">request</span><span class="kirari-punctuation">.</span><span class="kirari-property">url</span><span class="kirari-punctuation">)</span>
  <span class="kirari-keyword">const</span> <span class="kirari-variable">upstream</span> <span class="kirari-operator">=</span> <span class="kirari-function">routeByHosts</span><span class="kirari-punctuation">(</span><span class="kirari-variable">url</span><span class="kirari-punctuation">.</span><span class="kirari-property">hostname</span><span class="kirari-punctuation">)</span>
  <span class="kirari-keyword">if</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">upstream</span> <span class="kirari-operator">===</span> <span class="kirari-string">&#x27;&#x27;</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-keyword">return</span> <span class="kirari-keyword">new</span> <span class="kirari-function">Response</span><span class="kirari-punctuation">(</span>
      <span class="kirari-variable">JSON</span><span class="kirari-punctuation">.</span><span class="kirari-property">stringify</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">{</span>
        <span class="kirari-variable">routes</span><span class="kirari-punctuation">,</span>
      <span class="kirari-punctuation">}</span><span class="kirari-punctuation">)</span><span class="kirari-punctuation">,</span>
      <span class="kirari-punctuation">{</span>
        <span class="kirari-property">status</span><span class="kirari-punctuation">:</span> <span class="kirari-number">404</span><span class="kirari-punctuation">,</span>
      <span class="kirari-punctuation">}</span><span class="kirari-punctuation">,</span>
    <span class="kirari-punctuation">)</span>
  <span class="kirari-punctuation">}</span>
  <span class="kirari-keyword">const</span> <span class="kirari-variable">isDockerHub</span> <span class="kirari-operator">=</span> <span class="kirari-variable">upstream</span> <span class="kirari-operator">===</span> <span class="kirari-variable">dockerHub</span>
  <span class="kirari-keyword">const</span> <span class="kirari-variable">authorization</span> <span class="kirari-operator">=</span> <span class="kirari-variable">request</span><span class="kirari-punctuation">.</span><span class="kirari-property">headers</span><span class="kirari-punctuation">.</span><span class="kirari-property">get</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;Authorization&#x27;</span><span class="kirari-punctuation">)</span>
  <span class="kirari-keyword">if</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">url</span><span class="kirari-punctuation">.</span><span class="kirari-property">pathname</span> <span class="kirari-operator">===</span> <span class="kirari-string">&#x27;/v2/&#x27;</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-keyword">const</span> <span class="kirari-variable">newUrl</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">new</span> <span class="kirari-function">URL</span><span class="kirari-punctuation">(</span><span class="kirari-string">`</span><span class="kirari-punctuation">${</span><span class="kirari-variable">upstream</span><span class="kirari-punctuation">}</span><span class="kirari-string">/v2/`</span><span class="kirari-punctuation">)</span>
    <span class="kirari-keyword">const</span> <span class="kirari-variable">headers</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">new</span> <span class="kirari-function">Headers</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span>
    <span class="kirari-keyword">if</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">authorization</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
      <span class="kirari-variable">headers</span><span class="kirari-punctuation">.</span><span class="kirari-property">set</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;Authorization&#x27;</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">authorization</span><span class="kirari-punctuation">)</span>
    <span class="kirari-punctuation">}</span>
    <span class="kirari-comment">// check if need to authenticate</span>
    <span class="kirari-keyword">const</span> <span class="kirari-variable">resp</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">await</span> <span class="kirari-function">fetch</span><span class="kirari-punctuation">(</span><span class="kirari-variable">newUrl</span><span class="kirari-punctuation">.</span><span class="kirari-property">toString</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span><span class="kirari-punctuation">,</span> <span class="kirari-punctuation">{</span>
      <span class="kirari-property">method</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&#x27;GET&#x27;</span><span class="kirari-punctuation">,</span>
      <span class="kirari-variable">headers</span><span class="kirari-punctuation">,</span>
      <span class="kirari-property">redirect</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&#x27;follow&#x27;</span><span class="kirari-punctuation">,</span>
    <span class="kirari-punctuation">}</span><span class="kirari-punctuation">)</span>
    <span class="kirari-keyword">if</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">resp</span><span class="kirari-punctuation">.</span><span class="kirari-property">status</span> <span class="kirari-operator">===</span> <span class="kirari-number">401</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
      <span class="kirari-keyword">return</span> <span class="kirari-function">responseUnauthorized</span><span class="kirari-punctuation">(</span><span class="kirari-variable">url</span><span class="kirari-punctuation">)</span>
    <span class="kirari-punctuation">}</span>
    <span class="kirari-keyword">return</span> <span class="kirari-variable">resp</span>
  <span class="kirari-punctuation">}</span>
  <span class="kirari-comment">// get token</span>
  <span class="kirari-keyword">if</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">url</span><span class="kirari-punctuation">.</span><span class="kirari-property">pathname</span> <span class="kirari-operator">===</span> <span class="kirari-string">&#x27;/v2/auth&#x27;</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-keyword">const</span> <span class="kirari-variable">newUrl</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">new</span> <span class="kirari-function">URL</span><span class="kirari-punctuation">(</span><span class="kirari-string">`</span><span class="kirari-punctuation">${</span><span class="kirari-variable">upstream</span><span class="kirari-punctuation">}</span><span class="kirari-string">/v2/`</span><span class="kirari-punctuation">)</span>
    <span class="kirari-keyword">const</span> <span class="kirari-variable">resp</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">await</span> <span class="kirari-function">fetch</span><span class="kirari-punctuation">(</span><span class="kirari-variable">newUrl</span><span class="kirari-punctuation">.</span><span class="kirari-property">toString</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span><span class="kirari-punctuation">,</span> <span class="kirari-punctuation">{</span>
      <span class="kirari-property">method</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&#x27;GET&#x27;</span><span class="kirari-punctuation">,</span>
      <span class="kirari-property">redirect</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&#x27;follow&#x27;</span><span class="kirari-punctuation">,</span>
    <span class="kirari-punctuation">}</span><span class="kirari-punctuation">)</span>
    <span class="kirari-keyword">if</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">resp</span><span class="kirari-punctuation">.</span><span class="kirari-property">status</span> <span class="kirari-operator">!==</span> <span class="kirari-number">401</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
      <span class="kirari-keyword">return</span> <span class="kirari-variable">resp</span>
    <span class="kirari-punctuation">}</span>
    <span class="kirari-keyword">const</span> <span class="kirari-variable">authenticateStr</span> <span class="kirari-operator">=</span> <span class="kirari-variable">resp</span><span class="kirari-punctuation">.</span><span class="kirari-property">headers</span><span class="kirari-punctuation">.</span><span class="kirari-property">get</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;WWW-Authenticate&#x27;</span><span class="kirari-punctuation">)</span>
    <span class="kirari-keyword">if</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">authenticateStr</span> <span class="kirari-operator">===</span> <span class="kirari-null">null</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
      <span class="kirari-keyword">return</span> <span class="kirari-variable">resp</span>
    <span class="kirari-punctuation">}</span>
    <span class="kirari-keyword">const</span> <span class="kirari-variable">wwwAuthenticate</span> <span class="kirari-operator">=</span> <span class="kirari-function">parseAuthenticate</span><span class="kirari-punctuation">(</span><span class="kirari-variable">authenticateStr</span><span class="kirari-punctuation">)</span>
    <span class="kirari-keyword">let</span> <span class="kirari-variable">scope</span> <span class="kirari-operator">=</span> <span class="kirari-variable">url</span><span class="kirari-punctuation">.</span><span class="kirari-property">searchParams</span><span class="kirari-punctuation">.</span><span class="kirari-property">get</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;scope&#x27;</span><span class="kirari-punctuation">)</span>
    <span class="kirari-comment">// autocomplete repo part into scope for DockerHub library images</span>
    <span class="kirari-comment">// Example: repository:busybox:pull =&gt; repository:library/busybox:pull</span>
    <span class="kirari-keyword">if</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">scope</span> <span class="kirari-operator">&amp;&amp;</span> <span class="kirari-variable">isDockerHub</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
      <span class="kirari-keyword">const</span> <span class="kirari-variable">scopeParts</span> <span class="kirari-operator">=</span> <span class="kirari-variable">scope</span><span class="kirari-punctuation">.</span><span class="kirari-property">split</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;:&#x27;</span><span class="kirari-punctuation">)</span>
      <span class="kirari-keyword">if</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">scopeParts</span><span class="kirari-punctuation">.</span><span class="kirari-property">length</span> <span class="kirari-operator">===</span> <span class="kirari-number">3</span> <span class="kirari-operator">&amp;&amp;</span> <span class="kirari-operator">!</span><span class="kirari-variable">scopeParts</span><span class="kirari-punctuation">[</span><span class="kirari-number">1</span><span class="kirari-punctuation">]</span><span class="kirari-punctuation">.</span><span class="kirari-property">includes</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;/&#x27;</span><span class="kirari-punctuation">)</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
        <span class="kirari-variable">scopeParts</span><span class="kirari-punctuation">[</span><span class="kirari-number">1</span><span class="kirari-punctuation">]</span> <span class="kirari-operator">=</span> <span class="kirari-string">`library/</span><span class="kirari-punctuation">${</span><span class="kirari-variable">scopeParts</span><span class="kirari-punctuation">[</span><span class="kirari-number">1</span><span class="kirari-punctuation">]</span><span class="kirari-punctuation">}</span><span class="kirari-string">`</span>
        <span class="kirari-variable">scope</span> <span class="kirari-operator">=</span> <span class="kirari-variable">scopeParts</span><span class="kirari-punctuation">.</span><span class="kirari-property">join</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;:&#x27;</span><span class="kirari-punctuation">)</span>
      <span class="kirari-punctuation">}</span>
    <span class="kirari-punctuation">}</span>
    <span class="kirari-keyword">return</span> <span class="kirari-keyword">await</span> <span class="kirari-function">fetchToken</span><span class="kirari-punctuation">(</span><span class="kirari-variable">wwwAuthenticate</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">scope</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">authorization</span><span class="kirari-punctuation">)</span>
  <span class="kirari-punctuation">}</span>
  <span class="kirari-comment">// redirect for DockerHub library images</span>
  <span class="kirari-comment">// Example: /v2/busybox/manifests/latest =&gt; /v2/library/busybox/manifests/latest</span>
  <span class="kirari-keyword">if</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">isDockerHub</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-keyword">const</span> <span class="kirari-variable">pathParts</span> <span class="kirari-operator">=</span> <span class="kirari-variable">url</span><span class="kirari-punctuation">.</span><span class="kirari-property">pathname</span><span class="kirari-punctuation">.</span><span class="kirari-property">split</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;/&#x27;</span><span class="kirari-punctuation">)</span>
    <span class="kirari-keyword">if</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">pathParts</span><span class="kirari-punctuation">.</span><span class="kirari-property">length</span> <span class="kirari-operator">===</span> <span class="kirari-number">5</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
      <span class="kirari-variable">pathParts</span><span class="kirari-punctuation">.</span><span class="kirari-property">splice</span><span class="kirari-punctuation">(</span><span class="kirari-number">2</span><span class="kirari-punctuation">,</span> <span class="kirari-number">0</span><span class="kirari-punctuation">,</span> <span class="kirari-string">&#x27;library&#x27;</span><span class="kirari-punctuation">)</span>
      <span class="kirari-keyword">const</span> <span class="kirari-variable">redirectUrl</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">new</span> <span class="kirari-function">URL</span><span class="kirari-punctuation">(</span><span class="kirari-variable">url</span><span class="kirari-punctuation">)</span>
      <span class="kirari-variable">redirectUrl</span><span class="kirari-punctuation">.</span><span class="kirari-property">pathname</span> <span class="kirari-operator">=</span> <span class="kirari-variable">pathParts</span><span class="kirari-punctuation">.</span><span class="kirari-property">join</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;/&#x27;</span><span class="kirari-punctuation">)</span>
      <span class="kirari-keyword">return</span> <span class="kirari-variable">Response</span><span class="kirari-punctuation">.</span><span class="kirari-property">redirect</span><span class="kirari-punctuation">(</span><span class="kirari-variable">redirectUrl</span><span class="kirari-punctuation">,</span> <span class="kirari-number">301</span><span class="kirari-punctuation">)</span>
    <span class="kirari-punctuation">}</span>
  <span class="kirari-punctuation">}</span>
  <span class="kirari-comment">// foward requests</span>
  <span class="kirari-keyword">const</span> <span class="kirari-variable">newUrl</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">new</span> <span class="kirari-function">URL</span><span class="kirari-punctuation">(</span><span class="kirari-variable">upstream</span> <span class="kirari-operator">+</span> <span class="kirari-variable">url</span><span class="kirari-punctuation">.</span><span class="kirari-property">pathname</span><span class="kirari-punctuation">)</span>
  <span class="kirari-keyword">const</span> <span class="kirari-variable">newReq</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">new</span> <span class="kirari-function">Request</span><span class="kirari-punctuation">(</span><span class="kirari-variable">newUrl</span><span class="kirari-punctuation">,</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-property">method</span><span class="kirari-punctuation">:</span> <span class="kirari-variable">request</span><span class="kirari-punctuation">.</span><span class="kirari-property">method</span><span class="kirari-punctuation">,</span>
    <span class="kirari-property">headers</span><span class="kirari-punctuation">:</span> <span class="kirari-variable">request</span><span class="kirari-punctuation">.</span><span class="kirari-property">headers</span><span class="kirari-punctuation">,</span>
    <span class="kirari-property">redirect</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&#x27;follow&#x27;</span><span class="kirari-punctuation">,</span>
  <span class="kirari-punctuation">}</span><span class="kirari-punctuation">)</span>
  <span class="kirari-keyword">const</span> <span class="kirari-variable">resp</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">await</span> <span class="kirari-function">fetch</span><span class="kirari-punctuation">(</span><span class="kirari-variable">newReq</span><span class="kirari-punctuation">)</span>
  <span class="kirari-keyword">if</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">resp</span><span class="kirari-punctuation">.</span><span class="kirari-property">status</span> <span class="kirari-operator">===</span> <span class="kirari-number">401</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-keyword">return</span> <span class="kirari-function">responseUnauthorized</span><span class="kirari-punctuation">(</span><span class="kirari-variable">url</span><span class="kirari-punctuation">)</span>
  <span class="kirari-punctuation">}</span>
  <span class="kirari-keyword">return</span> <span class="kirari-variable">resp</span>
<span class="kirari-punctuation">}</span>

<span class="kirari-keyword">function</span> <span class="kirari-function">routeByHosts</span><span class="kirari-punctuation">(</span><span class="kirari-variable">host</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
  <span class="kirari-keyword">if</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">host</span> <span class="kirari-keyword">in</span> <span class="kirari-variable">routes</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-keyword">return</span> <span class="kirari-variable">routes</span><span class="kirari-punctuation">[</span><span class="kirari-variable">host</span><span class="kirari-punctuation">]</span>
  <span class="kirari-punctuation">}</span>
  <span class="kirari-keyword">if</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">MODE</span> <span class="kirari-operator">===</span> <span class="kirari-string">&#x27;debug&#x27;</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-keyword">return</span> <span class="kirari-variable">dockerHub</span>
  <span class="kirari-punctuation">}</span>
  <span class="kirari-keyword">return</span> <span class="kirari-string">&#x27;&#x27;</span>
<span class="kirari-punctuation">}</span>

<span class="kirari-keyword">function</span> <span class="kirari-function">parseAuthenticate</span><span class="kirari-punctuation">(</span><span class="kirari-variable">authenticateStr</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
  <span class="kirari-comment">// sample: Bearer realm=&quot;https://auth.ipv6.docker.com/token&quot;,service=&quot;registry.docker.io&quot;</span>
  <span class="kirari-comment">// match strings after =&quot; and before &quot;</span>
  <span class="kirari-keyword">const</span> <span class="kirari-variable">re</span> <span class="kirari-operator">=</span> <span class="kirari-regex">/(?&lt;==&quot;)(?:\\.|[^&quot;\\])*(?=&quot;)/g</span>
  <span class="kirari-keyword">const</span> <span class="kirari-variable">matches</span> <span class="kirari-operator">=</span> <span class="kirari-variable">authenticateStr</span><span class="kirari-punctuation">.</span><span class="kirari-property">match</span><span class="kirari-punctuation">(</span><span class="kirari-variable">re</span><span class="kirari-punctuation">)</span>
  <span class="kirari-keyword">if</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">matches</span> <span class="kirari-operator">==</span> <span class="kirari-null">null</span> <span class="kirari-operator">||</span> <span class="kirari-variable">matches</span><span class="kirari-punctuation">.</span><span class="kirari-property">length</span> <span class="kirari-operator">&lt;</span> <span class="kirari-number">2</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-keyword">throw</span> <span class="kirari-keyword">new</span> <span class="kirari-function">Error</span><span class="kirari-punctuation">(</span><span class="kirari-string">`invalid Www-Authenticate Header: </span><span class="kirari-punctuation">${</span><span class="kirari-variable">authenticateStr</span><span class="kirari-punctuation">}</span><span class="kirari-string">`</span><span class="kirari-punctuation">)</span>
  <span class="kirari-punctuation">}</span>
  <span class="kirari-keyword">return</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-property">realm</span><span class="kirari-punctuation">:</span> <span class="kirari-variable">matches</span><span class="kirari-punctuation">[</span><span class="kirari-number">0</span><span class="kirari-punctuation">]</span><span class="kirari-punctuation">,</span>
    <span class="kirari-property">service</span><span class="kirari-punctuation">:</span> <span class="kirari-variable">matches</span><span class="kirari-punctuation">[</span><span class="kirari-number">1</span><span class="kirari-punctuation">]</span><span class="kirari-punctuation">,</span>
  <span class="kirari-punctuation">}</span>
<span class="kirari-punctuation">}</span>

<span class="kirari-keyword">async</span> <span class="kirari-keyword">function</span> <span class="kirari-function">fetchToken</span><span class="kirari-punctuation">(</span><span class="kirari-variable">wwwAuthenticate</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">scope</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">authorization</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
  <span class="kirari-keyword">const</span> <span class="kirari-variable">url</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">new</span> <span class="kirari-function">URL</span><span class="kirari-punctuation">(</span><span class="kirari-variable">wwwAuthenticate</span><span class="kirari-punctuation">.</span><span class="kirari-property">realm</span><span class="kirari-punctuation">)</span>
  <span class="kirari-keyword">if</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">wwwAuthenticate</span><span class="kirari-punctuation">.</span><span class="kirari-property">service</span><span class="kirari-punctuation">.</span><span class="kirari-property">length</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-variable">url</span><span class="kirari-punctuation">.</span><span class="kirari-property">searchParams</span><span class="kirari-punctuation">.</span><span class="kirari-property">set</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;service&#x27;</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">wwwAuthenticate</span><span class="kirari-punctuation">.</span><span class="kirari-property">service</span><span class="kirari-punctuation">)</span>
  <span class="kirari-punctuation">}</span>
  <span class="kirari-keyword">if</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">scope</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-variable">url</span><span class="kirari-punctuation">.</span><span class="kirari-property">searchParams</span><span class="kirari-punctuation">.</span><span class="kirari-property">set</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;scope&#x27;</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">scope</span><span class="kirari-punctuation">)</span>
  <span class="kirari-punctuation">}</span>
  <span class="kirari-keyword">const</span> <span class="kirari-variable">headers</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">new</span> <span class="kirari-function">Headers</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span>
  <span class="kirari-keyword">if</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">authorization</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-variable">headers</span><span class="kirari-punctuation">.</span><span class="kirari-property">set</span><span class="kirari-punctuation">(</span><span class="kirari-string">&#x27;Authorization&#x27;</span><span class="kirari-punctuation">,</span> <span class="kirari-variable">authorization</span><span class="kirari-punctuation">)</span>
  <span class="kirari-punctuation">}</span>
  <span class="kirari-keyword">return</span> <span class="kirari-keyword">await</span> <span class="kirari-function">fetch</span><span class="kirari-punctuation">(</span><span class="kirari-variable">url</span><span class="kirari-punctuation">,</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-property">method</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&#x27;GET&#x27;</span><span class="kirari-punctuation">,</span>
    <span class="kirari-variable">headers</span>
  <span class="kirari-punctuation">}</span><span class="kirari-punctuation">)</span>
<span class="kirari-punctuation">}</span>

<span class="kirari-keyword">function</span> <span class="kirari-function">responseUnauthorized</span><span class="kirari-punctuation">(</span><span class="kirari-variable">url</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
  <span class="kirari-keyword">const</span> <span class="kirari-variable">headers</span> <span class="kirari-operator">=</span> <span class="kirari-keyword">new</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">Headers</span><span class="kirari-punctuation">)</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">)</span>
  <span class="kirari-keyword">if</span> <span class="kirari-punctuation">(</span><span class="kirari-variable">MODE</span> <span class="kirari-operator">===</span> <span class="kirari-string">&#x27;debug&#x27;</span><span class="kirari-punctuation">)</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-variable">headers</span><span class="kirari-punctuation">.</span><span class="kirari-property">set</span><span class="kirari-punctuation">(</span>
      <span class="kirari-string">&#x27;Www-Authenticate&#x27;</span><span class="kirari-punctuation">,</span>
      <span class="kirari-string">`Bearer realm=&quot;http://</span><span class="kirari-punctuation">${</span><span class="kirari-variable">url</span><span class="kirari-punctuation">.</span><span class="kirari-property">host</span><span class="kirari-punctuation">}</span><span class="kirari-string">/v2/auth&quot;,service=&quot;cloudflare-docker-proxy&quot;`</span><span class="kirari-punctuation">,</span>
    <span class="kirari-punctuation">)</span>
  <span class="kirari-punctuation">}</span>
  <span class="kirari-keyword">else</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-variable">headers</span><span class="kirari-punctuation">.</span><span class="kirari-property">set</span><span class="kirari-punctuation">(</span>
      <span class="kirari-string">&#x27;Www-Authenticate&#x27;</span><span class="kirari-punctuation">,</span>
      <span class="kirari-string">`Bearer realm=&quot;https://</span><span class="kirari-punctuation">${</span><span class="kirari-variable">url</span><span class="kirari-punctuation">.</span><span class="kirari-property">hostname</span><span class="kirari-punctuation">}</span><span class="kirari-string">/v2/auth&quot;,service=&quot;cloudflare-docker-proxy&quot;`</span><span class="kirari-punctuation">,</span>
    <span class="kirari-punctuation">)</span>
  <span class="kirari-punctuation">}</span>
  <span class="kirari-keyword">return</span> <span class="kirari-keyword">new</span> <span class="kirari-function">Response</span><span class="kirari-punctuation">(</span><span class="kirari-variable">JSON</span><span class="kirari-punctuation">.</span><span class="kirari-property">stringify</span><span class="kirari-punctuation">(</span><span class="kirari-punctuation">{</span>
    <span class="kirari-property">message</span><span class="kirari-punctuation">:</span> <span class="kirari-string">&#x27;UNAUTHORIZED&#x27;</span>
  <span class="kirari-punctuation">}</span><span class="kirari-punctuation">)</span><span class="kirari-punctuation">,</span> <span class="kirari-punctuation">{</span>
    <span class="kirari-property">status</span><span class="kirari-punctuation">:</span> <span class="kirari-number">401</span><span class="kirari-punctuation">,</span>
    <span class="kirari-variable">headers</span><span class="kirari-punctuation">,</span>
  <span class="kirari-punctuation">}</span><span class="kirari-punctuation">)</span>
<span class="kirari-punctuation">}</span>

<span class="kirari-keyword">export</span> <span class="kirari-keyword">default</span> <span class="kirari-punctuation">{</span>
  <span class="kirari-property">fetch</span><span class="kirari-punctuation">:</span> <span class="kirari-variable">handleRequest</span><span class="kirari-punctuation">,</span>
<span class="kirari-punctuation">}</span>
</code></pre><img src="https://c.statcounter.com/9823304/0/92a1d06c/1/" alt="" width="1" height="1" referrerPolicy="no-referrer-when-downgrade" />]]></content:encoded>
    </item>
  </channel>
</rss>