Codex检查网站死链怎么做：内部路径和外部状态要分开判断

<!doctype html><meta charset="utf-8"><title>入口</title>
<a href="docs/start.html">开始</a>

<!doctype html><meta charset="utf-8"><title>开始</title>
<a href="../guide.html#setup">正确锚点</a>
<a href="../guide.html#gone">不存在的锚点</a>
<a href="/guide">重写路由</a>
<a href="../missing.html">缺失页面</a>
<a href="https://outside.example/article">外部待查</a>

<!doctype html><meta charset="utf-8"><title>指南</title>
<h1 id="setup">开始配置</h1>

import fs from 'node:fs';
import path from 'node:path';
import {chromium} from 'playwright';
const root=path.resolve('public'),origin='https://example.test';
const routes={'/guide':'guide.html'};
const sources=['index.html','docs/start.html','guide.html'];
const browser=await chromium.launch({
 headless:true,...(process.env.PW_CHANNEL?{channel:process.env.PW_CHANNEL}:{})
});
const page=await browser.newPage(),report=[];
try{
 for(const source of sources){
  const html=fs.readFileSync(path.join(root,source),'utf8');
  const hrefs=await page.evaluate(text=>{
   const doc=new DOMParser().parseFromString(text,'text/html');
   return [...doc.querySelectorAll('a[href]')].map(a=>({raw:a.getAttribute('href'),label:a.textContent}));
  },html);
  for(const link of hrefs){
   const row={source,...link};
   let url;try{url=new URL(link.raw,origin+'/'+source);}catch{report.push({...row,status:'invalid-url'});continue;}
   if(!['http:','https:'].includes(url.protocol)){report.push({...row,status:'non-http'});continue;}
   row.resolved=url.href;
   if(url.origin!==origin){report.push({...row,status:'external-not-probed'});continue;}
   let decoded;try{decoded=decodeURIComponent(url.pathname);}catch{report.push({...row,status:'invalid-encoding'});continue;}
   const relative=routes[url.pathname]??decoded.replace(/^\/+/,'');
   const target=path.resolve(root,relative),rel=path.relative(root,target);
   if(rel==='..'||rel.startsWith('..'+path.sep)||path.isAbsolute(rel)){
    report.push({...row,status:'out-of-scope'});continue;
   }
   if(!fs.existsSync(target)||!fs.statSync(target).isFile()){
    report.push({...row,target:rel,status:'missing-file'});continue;
   }
   if(url.hash){
    let id;try{id=decodeURIComponent(url.hash.slice(1));}catch{report.push({...row,status:'invalid-fragment'});continue;}
    const found=await page.evaluate(({text,id})=>{
     const doc=new DOMParser().parseFromString(text,'text/html');
     return !!doc.getElementById(id);
    },{text:fs.readFileSync(target,'utf8'),id});
    if(!found){report.push({...row,target:rel,status:'missing-fragment'});continue;}
   }
   report.push({...row,target:rel,status:routes[url.pathname]?'route-mapped':'local-ok'});
  }
 }
}finally{await browser.close();}
console.log(JSON.stringify(report,null,2));

运行小样扫描，保留每条来源与目标。
把 #gone 改到实际 setup；缺失页面链接根据上下文改为 guide.html。
只改教学源 public/docs/start.html，其他 href 保留；重扫后比较新旧报告。
预期 missing-file 与 missing-fragment 各少一条，其他类别不被吞掉。
真实生成站要先追到链接的内容源再改，不只改输出 HTML。
外链本轮不访问；需要外部探测时另列允许域名、超时与低并发范围。