fix(viral-video): xml_parser 剥离 CDATA 包裹，修复 copy_display_markdown 前端泄露 (#2238)

Co-authored-by: xiaoxia <dev@xiaoxiajianji.com>
Co-committed-by: xiaoxia <dev@xiaoxiajianji.com>

packages/application/viral_video/xml_parser.py

@@ -13,6 +13,13 @@ from typing import Optional
 _OPEN_RE = re.compile(r"<(?P<tag>[\w-]+)(?P<attrs>(?:\s(?:[^>]*?\S)?)?)(?P<self>/?)>")
 _CLOSE_RE = re.compile(r"</(?P<tag>[\w-]+)\s*>")
 _ATTR_RE = re.compile(r"""([\w:-]+)\s*=\s*(?:"([^"]*)"|'([^']*)')""")
+_CDATA_RE = re.compile(r"^<!\[CDATA\[(.*)\]\]>$", re.DOTALL)
+
+
+def _strip_cdata(s: str) -> str:
+    """剥离 LLM 可能照抄示例输出的 ``<![CDATA[...]]>`` 包裹层。"""
+    m = _CDATA_RE.match(s.strip())
+    return m.group(1) if m else s
 
 
 def parse_attributes(raw: str) -> dict[str, str]:
@@ -58,6 +65,7 @@ def parse_tags(text: Optional[str]) -> list[dict]:
                 if stack[idx]["tag"] == tag:
                     node = stack[idx]
                     node["text"] = unescape(text[node["_start"] : token.start()].strip())
+                    node["text"] = _strip_cdata(node["text"])
                     node.pop("_start", None)
                     del stack[idx:]
                     break
@@ -65,6 +73,7 @@ def parse_tags(text: Optional[str]) -> list[dict]:
     for node in stack:
         if "_start" in node:
             node["text"] = unescape(text[node["_start"] :].strip())
+            node["text"] = _strip_cdata(node["text"])
             node.pop("_start", None)
     return results
 

tests/unit/test_xml_parser.py

@@ -0,0 +1,49 @@
+"""xml_parser CDATA 剥离单元测试。"""
+
+from packages.application.viral_video.xml_parser import find_all, text_of
+
+XML = """<script>
+<copy_display_markdown><![CDATA[# 标题
+
+这是第一段，含**加粗**和[链接](https://a.com)。
+
+第二行，保留换行。]]></copy_display_markdown>
+<voiceover>口播不带 CDATA，保持原样。</voiceover>
+<visual><![CDATA[画面：产品特写，光线柔和]]></visual>
+<action_details><![CDATA[未闭合标签里的 CDATA 也要剥离]]></action_details>
+</script>"""
+
+
+def test_text_of_strips_cdata_with_markdown_newlines():
+    text = text_of(XML, "copy_display_markdown")
+    assert not text.startswith("<![CDATA[")
+    assert not text.endswith("]]>")
+    assert "# 标题" in text
+    assert "**加粗**" in text
+    assert "[链接](https://a.com)" in text
+    # markdown 换行被保留
+    assert "\n\n第二行" in text
+
+
+def test_plain_text_unchanged():
+    assert text_of(XML, "voiceover") == "口播不带 CDATA，保持原样。"
+
+
+def test_other_cdata_fields_stripped():
+    assert text_of(XML, "visual") == "画面：产品特写，光线柔和"
+
+
+def test_unclosed_tag_cdata_stripped():
+    # action_details 没有闭合标签，走未闭合兜底分支
+    node = find_all(XML, "action_details")[0]
+    assert node["text"] == "未闭合标签里的 CDATA 也要剥离"
+
+
+def test_no_cdata_returns_original():
+    xml = "<copy_display_markdown>普通内容]]> 残留结尾</copy_display_markdown>"
+    # 非完整 CDATA 包裹不应被误剥离
+    assert text_of(xml, "copy_display_markdown") == "普通内容]]> 残留结尾"
+
+
+def test_missing_tag_default():
+    assert text_of(XML, "nope", default="缺省") == "缺省"