feat:自动剥离大模型返回的markdown代码围栏

backend/app/routers/html.py

@@ -48,6 +48,12 @@ CONTENT_SECURITY_POLICY = "; ".join(
     ]
 )
 
+MARKDOWN_CODE_FENCE_PATTERN = re.compile(
+    r"^\s*```(?P<language>[a-zA-Z0-9_+-]*)[^\S\r\n]*\r?\n"
+    r"(?P<content>[\s\S]*?)"
+    r"\r?\n```\s*$"
+)
+
 
 def build_response_headers() -> dict[str, str]:
     headers = {
@@ -61,6 +67,32 @@ def build_response_headers() -> dict[str, str]:
     return headers
 
 
+def strip_outer_markdown_code_fence(
+    raw_content: str,
+    preserve_markdown_fence: bool,
+) -> str:
+    if preserve_markdown_fence or not settings.strip_markdown_code_fence:
+        return raw_content
+
+    match = MARKDOWN_CODE_FENCE_PATTERN.match(raw_content)
+    if not match:
+        return raw_content
+
+    language = match.group("language").strip().lower()
+    content = match.group("content")
+    looks_like_html = bool(
+        re.search(r"<!doctype\s+html|<html\b|<[a-z][\w:-]*\b", content, re.IGNORECASE)
+    )
+
+    if language and language not in {"html", "htm", "xhtml", "xml"}:
+        return raw_content
+
+    if not language and not looks_like_html:
+        return raw_content
+
+    return content
+
+
 def require_api_key(x_api_key: str | None = Header(default=None, alias="X-API-Key")) -> None:
     if not settings.api_key:
         return
@@ -261,7 +293,11 @@ def generate_html(
         unique_id = generate_unique_id(db)
         html_filename = f"{unique_id}.html"
         html_path = settings.html_storage_dir / html_filename
-        html_document = build_html_document(request.html_content, request.title)
+        normalized_html_content = strip_outer_markdown_code_fence(
+            request.html_content,
+            preserve_markdown_fence=request.preserve_markdown_fence,
+        )
+        html_document = build_html_document(normalized_html_content, request.title)
         expires_at = datetime.utcnow() + timedelta(
             days=request.ttl_days or settings.default_retention_days
         )