Playwright Request Guard

1'use strict';
2// Preloaded (`node --require`) into the design-md-generator process. That generator is a separate CLI that
3// launches its own Chromium and navigates to the source URL, so it cannot call the in-process guard. This hook
4// wraps Playwright's launch so every context/page it creates gets the same request guard as the other
5// extractors: any navigation, redirect hop or subresource to a non-public destination is aborted.
6const Module = require('node:module');
7const path = require('node:path');
8const { pathToFileURL } = require('node:url');
9
10const guardModule = import(pathToFileURL(path.join(__dirname, 'url-safety.mjs')).href);
11const guarded = new WeakSet();
12
13async function guard(target) {
14  if (guarded.has(target)) return target;
15  guarded.add(target);
16  const { guardPageRequests } = await guardModule;
17  await guardPageRequests(target);
18  return target;
19}
20
21function wrapBrowser(browser) {
22  const newContext = browser.newContext.bind(browser);
23  const newPage = browser.newPage.bind(browser);
24  browser.newContext = async (...args) => guard(await newContext(...args));
25  browser.newPage = async (...args) => guard(await newPage(...args));
26  return browser;
27}
28
29function wrapChromium(chromium) {
30  if (!chromium || chromium.__rpRequestGuard) return;
31  chromium.__rpRequestGuard = true;
32  const launch = chromium.launch.bind(chromium);
33  chromium.launch = async (...args) => wrapBrowser(await launch(...args));
34  if (chromium.launchPersistentContext) {
35    const persistent = chromium.launchPersistentContext.bind(chromium);
36    chromium.launchPersistentContext = async (...args) => guard(await persistent(...args));
37  }
38  if (chromium.connectOverCDP) {
39    const connect = chromium.connectOverCDP.bind(chromium);
40    chromium.connectOverCDP = async (...args) => wrapBrowser(await connect(...args));
41  }
42}
43
44// Hook Module._load rather than Module.prototype.require: it is what `require()` calls, and also what the
45// ESM loader calls when an ES module `import`s a CommonJS package such as playwright, so both CLI styles of
46// the generator are covered. The ESM entry resolves to an absolute path, hence the substring match.
47const originalLoad = Module._load;
48Module._load = function patchedLoad(request) {
49  const loaded = originalLoad.apply(this, arguments);
50  if (typeof request === 'string' && request.includes('playwright')) {
51    try {
52      wrapChromium(loaded && loaded.chromium);
53    } catch {
54      /* an export that is not Playwright's */
55    }
56  }
57  return loaded;
58};