[colinodell]

Fixes #254

HtmlConverter::sanitize() decodes the serialized document before stripping the document's own wrapper tags from it, so text inside of code which looks like one of those tags, once decoded, could be mistaken for it:

Input	Before	After
<pre><code>&lt;/body&gt;</code></pre>	```\n</body>\n```\n\n</body>	```\n</body>\n```
<p>a <code>&lt;/body&gt;&lt;/html&gt;</code></p>	a `</body></html>`</body></html>	a `</body></html>`
<p>Use <code>&lt;!DOCTYPE html</code> first.</p><p>a &gt; b</p>	Use `	Use `<!DOCTYPE html` first.\n\na &gt; b

• The closing tags were found with strpos(), which returns the first occurrence, so an escaped one in code was taken for the real one and the real tags at the end were left in. They're now only stripped where they end the output.
• The doctype regex wasn't anchored, so it removed everything from an escaped <!DOCTYPE in code up to the next >, which could be the rest of the document. saveHTML() always puts the real doctype first (checked on libxml 2.9.4, 2.9.14 and 2.13.9, including documents with no doctype, a lowercase one, or comments and processing instructions before it), so it's now anchored to the start.

Whatever the old checks stripped, the new ones still strip; they only stop stripping the copies which aren't at the start or end. Preserved comments containing </body> are fixed by the same change.

This deliberately keeps the change to sanitize() minimal, as it's a protected method and its input is unchanged. A wider rework which avoids serializing and decoding the document altogether (which would also clean up leftovers like </head><body> from documents with a <head>) is better suited to a minor release.

🤖 Generated with Claude Code

https://claude.ai/code/session_0192z2KwkS1QtyN2hrGVpQ4z

[Copilot]

🟢 Approval recommended

The narrowly scoped changes address the reported failures with targeted regression coverage and no unresolved findings.

No open issues 🎉

What changed in this PR

Fixes #254 by preventing escaped document tags in code and preserved comments from interfering with HTML-to-Markdown wrapper removal.

Changes:

• Restricts doctype removal to the start and closing-tag removal to the end.
• Adds regression tests for document-like text in code and comments.
• Documents the fix in the changelog.

File	Description
tests/​HtmlConverterTest.php	Adds regression coverage for escaped document tags.
src/​HtmlConverter.php	Anchors document wrapper removal to output boundaries.
CHANGELOG.md	Records the fixes for #254.

_{Review effort: Balanced}