style: auto-format with black + isort + ruff + prettier [skip ci-format-check]

P0: 前端刷新或WS重连时，initial消息只有status，导致imageAnalysis state为null、识别描述区域不渲染。
修复：initial.data里按状态对齐worker事件结构，带上image_analysis/copy_result/storyboard/generated_copy_text；
copy_result复用已有_build_copy_result(job)（兼容v1.5老数据）；中间态无业务字段时不塞空dict/list。
新增3个单测覆盖 image_analyzed / copy_generated / 中间态纯净快照。

apps/api/app/api/routes/viral_video.py

@@ -890,13 +890,28 @@ async def viral_video_websocket(websocket: WebSocket, job_id: str) -> None:
         job = job_repo.get(job_id)
         if job is not None:
             status_val = job.status.value if hasattr(job.status, "value") else str(job.status)
+            # #P0: 初始快照必须包含前端重连/刷新所需的业务字段，
+            # 结构对齐 worker 推送的 image_analyzed / copy_generated 事件。
+            data: dict = {"status": status_val}
+            ia = getattr(job, "image_analysis", None)
+            if isinstance(ia, dict) and ia:
+                data["image_analysis"] = ia
+            cr = _build_copy_result(job)
+            if isinstance(cr, dict) and cr:
+                data["copy_result"] = cr
+            gct = getattr(job, "generated_copy_text", "") or ""
+            if gct:
+                data["generated_copy_text"] = gct
+            sb = getattr(job, "storyboard", None) or []
+            if sb:
+                data["storyboard"] = sb
             initial = {
                 "type": "viral_video:progress",
                 "job_id": job_id,
                 "stage": _stage_from_status(job),
                 "progress": _estimate_progress(job),
                 "message": _initial_message(job),
-                "data": {"status": status_val},
+                "data": data,
             }
             await websocket.send_json(initial)
             # 已经终态 → 再发一条终态事件后立即关闭，避免占连接

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_viral_video_ws.py

@@ -430,3 +430,56 @@ class TestWSInitialSnapshot:
             for p in patches:
                 p.stop()
         assert any(m["type"] == "forwarder_reached" for m in received)
+
+    def test_image_analyzed_initial_snapshot_contains_image_analysis(self):
+        """P0: image_analyzed 状态时初始快照必须带 image_analysis。"""
+        ia = {"images": [{"type": "product", "name": "X", "summary_markdown": "# X\nhello"}]}
+        job = _make_job(
+            status="image_analyzed",
+            user_id="user-a",
+            is_terminal=False,
+            image_analysis=ia,
+            copy_result=None,
+            generated_copy_text="",
+            storyboard=[],
+        )
+        received, _, _ = _run_ws_handshake(job=job)
+        assert received[0]["data"]["status"] == "image_analyzed"
+        assert received[0]["data"]["image_analysis"] == ia
+
+    def test_copy_generated_initial_snapshot_contains_copy_result(self):
+        """P0: copy_generated 状态时初始快照必须带 copy_result/storyboard。"""
+        cr = {"shots": [{"time_range": "0-3s", "voiceover": "hi"}], "voiceover_script": "hi"}
+        sb = [{"order": 1, "text": "hi", "duration": 3.0}]
+        job = _make_job(
+            status="copy_generated",
+            user_id="user-a",
+            is_terminal=False,
+            image_analysis={"images": []},
+            copy_result=cr,
+            generated_copy_text="hi",
+            storyboard=sb,
+        )
+        received, _, _ = _run_ws_handshake(job=job)
+        data = received[0]["data"]
+        assert data["status"] == "copy_generated"
+        # _build_copy_result 会补 final_copy/suggested_copy/title 兜底
+        assert data["copy_result"]["shots"] == cr["shots"]
+        assert data["storyboard"] == sb
+        assert data["generated_copy_text"] == "hi"
+        assert data["image_analysis"] == {"images": []}
+
+    def test_initial_snapshot_without_business_fields_only_has_status(self):
+        """running/pending 等中间态，无业务字段时不应塞空 dict/list。"""
+        job = _make_job(
+            status="running",
+            user_id="user-a",
+            is_terminal=False,
+            image_analysis=None,
+            copy_result=None,
+            generated_copy_text="",
+            storyboard=[],
+        )
+        received, _, _ = _run_ws_handshake(job=job)
+        data = received[0]["data"]
+        assert data == {"status": "running"}

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"