feat:自动剥离大模型返回的markdown代码围栏

backend/.env

@@ -8,5 +8,6 @@ MAX_RETENTION_DAYS=30
 MAX_HTML_LENGTH=200000
 API_KEY=""
 ALLOW_UNSAFE_HTML=true
+STRIP_MARKDOWN_CODE_FENCE=true
 ENABLE_REQUEST_DEBUG_LOG=true
 REQUEST_LOG_MAX_CHARS=10000

backend/.env.example

@@ -8,5 +8,6 @@ MAX_RETENTION_DAYS=30
 MAX_HTML_LENGTH=200000
 API_KEY=""
 ALLOW_UNSAFE_HTML=true
+STRIP_MARKDOWN_CODE_FENCE=true
 ENABLE_REQUEST_DEBUG_LOG=true
 REQUEST_LOG_MAX_CHARS=10000

backend/app/config.py

@@ -76,6 +76,7 @@ class Settings:
     max_html_length = max(1024, _get_int_env("MAX_HTML_LENGTH", 200_000))
     api_key = os.getenv("API_KEY", "").strip()
     allow_unsafe_html = _get_bool_env("ALLOW_UNSAFE_HTML", False)
+    strip_markdown_code_fence = _get_bool_env("STRIP_MARKDOWN_CODE_FENCE", True)
     enable_request_debug_log = _get_bool_env("ENABLE_REQUEST_DEBUG_LOG", True)
     request_log_max_chars = max(256, _get_int_env("REQUEST_LOG_MAX_CHARS", 10_000))
 

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
         )

backend/app/schemas.py

@@ -30,6 +30,13 @@ class HTMLGenerateRequest(BaseModel):
         ge=1,
         description="Optional retention days for the file.",
     )
+    preserve_markdown_fence: bool = Field(
+        default=False,
+        description=(
+            "When true, keeps outer Markdown code fences instead of stripping them "
+            "from the HTML payload."
+        ),
+    )
 
     @root_validator(pre=True)
     def normalize_aliases(cls, values: dict) -> dict:
@@ -73,6 +80,17 @@ class HTMLGenerateRequest(BaseModel):
         normalized = value.strip()
         return normalized or None
 
+    @validator("preserve_markdown_fence", pre=True)
+    def normalize_preserve_markdown_fence(cls, value: bool | str | None) -> bool:
+        if isinstance(value, bool):
+            return value
+        if value is None:
+            return False
+        if isinstance(value, str):
+            return value.strip().lower() in {"1", "true", "yes", "on"}
+
+        return bool(value)
+
     @validator("ttl_days")
     def validate_ttl_days(cls, value: int | None) -> int | None:
         if value is None: