From f6a5758b8ce48e6ba9decbf26484257b63623545 Mon Sep 17 00:00:00 2001
From: Andrew Chen <48723787+chuenchen309@users.noreply.github.com>
Date: Thu, 16 Jul 2026 00:28:46 +0800
Subject: [PATCH] fix(html2text): thread escape_backslash into
 escape_md_section() call

Sibling flags (escape_dot/escape_plus/escape_dash) are all forwarded from
the HTML2Text instance into escape_md_section() at its call site in
handle_data(), but escape_backslash was never passed. As a result,
escape_md_section()'s own hardcoded default of True always wins, so
backslashes are escaped/doubled in generated markdown regardless of what
the instance (or CustomHTML2Text's own escape_backslash=False default,
used by the default markdown-generation pipeline) configures.

Fix: pass escape_backslash=self.escape_backslash alongside the other
escape_* flags, the same way the siblings are threaded.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 crawl4ai/html2text/__init__.py                |  1 +
 tests/unit/test_html2text_escape_backslash.py | 35 +++++++++++++++++++
 2 files changed, 36 insertions(+)
 create mode 100644 tests/unit/test_html2text_escape_backslash.py

diff --git a/crawl4ai/html2text/__init__.py b/crawl4ai/html2text/__init__.py
index ff39fb211..2b77ea865 100644
--- a/crawl4ai/html2text/__init__.py
+++ b/crawl4ai/html2text/__init__.py
@@ -952,6 +952,7 @@ def handle_data(self, data: str, entity_char: bool = False) -> None:
             data = escape_md_section(
                 data,
                 snob=self.escape_snob,
+                escape_backslash=self.escape_backslash,
                 escape_dot=self.escape_dot,
                 escape_plus=self.escape_plus,
                 escape_dash=self.escape_dash,
diff --git a/tests/unit/test_html2text_escape_backslash.py b/tests/unit/test_html2text_escape_backslash.py
new file mode 100644
index 000000000..7f15092cc
--- /dev/null
+++ b/tests/unit/test_html2text_escape_backslash.py
@@ -0,0 +1,35 @@
+"""Unit tests for CustomHTML2Text's escape_backslash config flag.
+
+escape_backslash is threaded from config.py -> HTML2Text.__init__ -> instance
+attribute, and sibling flags (escape_dot/escape_plus/escape_dash) are all
+forwarded into escape_md_section() at its call site in handle_data(). This
+tests that escape_backslash is forwarded the same way, so that setting it to
+False actually disables backslash escaping in the generated markdown.
+"""
+from crawl4ai.html2text import CustomHTML2Text
+
+
+def _convert(html: str, escape_backslash: bool) -> str:
+    h = CustomHTML2Text()
+    h.body_width = 0
+    h.escape_backslash = escape_backslash
+    return h.handle(html)
+
+
+# RE_MD_BACKSLASH_MATCHER (html2text/config.py) only fires on a backslash
+# followed by a markdown-special char (`*_{}[]()#+-.!), so the repro text
+# must contain a literal backslash immediately before one of those chars.
+HTML_WITH_BACKSLASH = r"<p>Use \* for multiplication</p>"
+
+
+class TestEscapeBackslashConfig:
+    def test_escape_backslash_false_leaves_backslashes_alone(self):
+        """escape_backslash=False must not double the literal backslash."""
+        out = _convert(HTML_WITH_BACKSLASH, escape_backslash=False)
+        assert r"\*" in out
+        assert r"\\*" not in out
+
+    def test_escape_backslash_true_still_escapes(self):
+        """escape_backslash=True (the escape_md_section default) must still escape."""
+        out = _convert(HTML_WITH_BACKSLASH, escape_backslash=True)
+        assert r"\\*" in out