defuddle vulnerable to XSS via unescaped string interpolation in _findContentBySchemaText image tag

Summary

The _findContentBySchemaText method in src/defuddle.ts interpolates image src and alt attributes directly into an HTML string without escaping:

html += `<img src="${imageSrc}" alt="${imageAlt}">`;

An attacker can use a " in the alt attribute to break out of the attribute context and inject event handlers. This is a separate vulnerability from the sanitization bypass fixed in f154cb7 — the injection happens during string construction, not in the DOM, so _stripUnsafeElements cannot catch it.

Details

When _findContentBySchemaText finds a sibling image outside the matched content element, it reads the image's src and alt attributes via getAttribute() and interpolates them into a template literal. getAttribute('alt') returns the raw attribute value. If the alt contains ", it terminates the alt attribute in the interpolated HTML string, and subsequent content becomes new attributes (including event handlers).

The recently added _stripUnsafeElements() (commit f154cb7) strips on* attributes from DOM elements, but the alt attribute's name is alt (not on*), so it is preserved with its full value. The onload handler is created by the string interpolation, not present in the original DOM.

PoC

Input HTML:

<!DOCTYPE html>
<html>
<head>
<title>PoC</title>
<script type="application/ld+json">
{"@type": "Article", "text": "Long article text repeated many times to exceed the extracted content word count. Long article text repeated many times to exceed the extracted content word count. Long article text repeated many times to exceed the extracted content word count."}
</script>
</head>
<body>
<article><p>Short.</p></article>
<div class="post-container">
  <p>Extra text to inflate parent word count padding padding padding.</p>
  <div class="post-body">
    Long article text repeated many times to exceed the extracted content word count. Long article text repeated many times to exceed the extracted content word count. Long article text repeated many times to exceed the extracted content word count.
  </div>
  <img width="800" height="600" src="https://example.com/photo.jpg" alt='pwned" onload="alert(document.cookie)'>
</div>
</body>
</html>

Output:

<img src="https://example.com/photo.jpg" alt="pwned" onload="alert(document.cookie)">

The onload event handler is injected as a separate HTML attribute.

Impact

XSS in any application that renders defuddle's HTML output (browser extensions, web clippers, reader modes). The attack requires crafted HTML with schema.org structured data that triggers the _findContentBySchemaText fallback, combined with a sibling image whose alt attribute contains a quote character followed by an event handler.

Suggested Fix

Use DOM API instead of string interpolation:

if (imageSrc) {
    const img = this.doc.createElement('img');
    img.setAttribute('src', imageSrc);
    img.setAttribute('alt', imageAlt);
    html += img.outerHTML;
}

This ensures attribute values are properly escaped by the DOM serializer.