fix: 修复报告摘要生成并稳定默认端口

app/platforms/douyin.py

@@ -88,11 +88,23 @@ class DouyinPlatform:
         return self.map_items(payload, limit=limit)
 
     def fetch_comments(self, source_item_id: str, *, limit: int) -> list[CommentData]:
-        payload = self.client.get(
+        comments: list[CommentData] = []
-            "/api/v1/douyin/app/v3/fetch_video_comments",
+        cursor = 0
-            params={"aweme_id": source_item_id, "cursor": 0, "count": 20},
+        page_size = min(20, limit)
-        )
+        while len(comments) < limit:
-        return self.map_comments(payload, limit=limit)
+            payload = self.client.get(
+                "/api/v1/douyin/app/v3/fetch_video_comments",
+                params={"aweme_id": source_item_id, "cursor": cursor, "count": page_size},
+            )
+            page_comments = self.map_comments(payload, limit=limit - len(comments))
+            comments.extend(page_comments)
+            data = payload.get("data", {})
+            next_cursor = data.get("cursor") or payload.get("cursor")
+            has_more = data.get("has_more", payload.get("has_more", 0))
+            if not page_comments or not has_more or next_cursor in (None, cursor):
+                break
+            cursor = next_cursor
+        return comments[:limit]
 
     def _author_name(self, comment: dict[str, Any]) -> str | None:
         user = comment.get("user") or {}

app/services/ai_service.py

@@ -42,6 +42,46 @@ def build_comment_prompt(comments: list[dict[str, str]]) -> str:
     )
 
 
+def build_report_summary_prompt(metrics: dict, typical: dict, *, word_limit: int) -> str:
+    sentiment = metrics.get("sentiment", {})
+    sentiment_lines = []
+    for name in ("positive", "neutral", "negative", "unknown"):
+        entry = sentiment.get(name, {})
+        sentiment_lines.append(f"- {name}: {entry.get('count', 0)} ({entry.get('pct', 0)}%)")
+
+    label_lines = [
+        f"- {label.get('name', '')}: {label.get('count', 0)}"
+        for label in metrics.get("top_labels", [])
+        if label.get("name")
+    ]
+    if not label_lines:
+        label_lines = ["- 暂无"]
+
+    typical_lines = []
+    for sentiment_name, comments in typical.items():
+        for comment in comments:
+            typical_lines.append(f"- [{sentiment_name}] {str(comment.get('content', ''))[:150]}")
+    if not typical_lines:
+        typical_lines = ["- 暂无"]
+
+    item_count_line = ""
+    if "item_count" in metrics:
+        item_count_line = f"内容条目数量：{metrics.get('item_count', 0)}\n"
+
+    return (
+        f"只返回一段中文总结，不使用 Markdown，不超过 {word_limit} 字。\n"
+        "总结必须基于以下统计数据和典型评论，不要编造未提供的信息。\n"
+        f"{item_count_line}"
+        f"样本评论数量：{metrics.get('sample_count', 0)}\n"
+        "情绪分布：\n"
+        f"{chr(10).join(sentiment_lines)}\n"
+        "Top 标签：\n"
+        f"{chr(10).join(label_lines)}\n"
+        "典型评论：\n"
+        f"{chr(10).join(typical_lines)}"
+    )
+
+
 def parse_ai_comment_response(response_text: str, *, expected_comment_ids: set[str]) -> list[AIAnalysisResult]:
     try:
         raw = json.loads(response_text)
@@ -118,7 +158,7 @@ class OpenAICompatibleAIClient:
         self.model = model
         self._http_client = http_client or httpx.Client(timeout=timeout_seconds)
 
-    def request(self, prompt: str) -> str:
+    def request(self, prompt: str, *, system_prompt: str = "请严格遵循用户指令输出。") -> str:
         endpoint = f"{self.base_url}/chat/completions" if self.base_url.endswith("/v1") else f"{self.base_url}/v1/chat/completions"
         response = self._http_client.post(
             endpoint,
@@ -128,7 +168,7 @@ class OpenAICompatibleAIClient:
                 "messages": [
                     {
                         "role": "system",
-                        "content": "只返回 JSON Array，不输出 Markdown 或解释性自然语言。",
+                        "content": system_prompt,
                     },
                     {"role": "user", "content": prompt},
                 ],

app/services/report_service.py

@@ -128,7 +128,12 @@ def _safe_summary(summary_provider, metrics: dict, typical: dict, *, limit: int)
     if summary_provider is None:
         return DEFAULT_SUMMARY
     try:
-        summary = summary_provider(metrics, typical)
+        summary = summary_provider(metrics, typical, word_limit=limit)
+    except TypeError:
+        try:
+            summary = summary_provider(metrics, typical)
+        except Exception:
+            return DEFAULT_SUMMARY
     except Exception:
         return DEFAULT_SUMMARY
     return str(summary)[:limit]

app/services/task_service.py

@@ -13,7 +13,7 @@ from app.platforms.base import PlatformAPIError, TikHubClient
 from app.platforms.douyin import DouyinPlatform
 from app.platforms.xiaohongshu import XiaohongshuPlatform
 from app.schemas import CreateTaskRequest
-from app.services.ai_service import OpenAICompatibleAIClient, analyze_comments_with_retry, calculate_analysis_status
+from app.services.ai_service import OpenAICompatibleAIClient, analyze_comments_with_retry, build_report_summary_prompt, calculate_analysis_status
 from app.services.report_service import generate_hotspot_report, generate_item_report
 
 
@@ -52,16 +52,25 @@ def build_platform(platform: str):
     raise ValueError(f"Unsupported platform: {platform}")
 
 
-def build_ai_requester():
+def build_ai_client() -> OpenAICompatibleAIClient | None:
     settings = get_settings()
     if settings.ai_base_url and settings.ai_api_key and settings.ai_model:
-        client = OpenAICompatibleAIClient(
+        return OpenAICompatibleAIClient(
             base_url=settings.ai_base_url,
             api_key=settings.ai_api_key,
             model=settings.ai_model,
             timeout_seconds=settings.ai_timeout_seconds,
         )
-        return client.request
+    return None
+
+
+def build_ai_requester(client: OpenAICompatibleAIClient | None = None):
+    client = client if client is not None else build_ai_client()
+    if client is not None:
+        return lambda prompt: client.request(
+            prompt,
+            system_prompt="只返回 JSON Array，不输出 Markdown 或解释性自然语言。",
+        )
 
     def fallback_requester(_prompt: str) -> str:
         return "[]"
@@ -69,6 +78,25 @@ def build_ai_requester():
     return fallback_requester
 
 
+def build_report_summary_provider(client: OpenAICompatibleAIClient | None):
+    if client is None:
+        return None
+
+    def summarize_report(metrics: dict, typical: dict, *, word_limit: int = 200) -> str:
+        return client.request(
+            build_report_summary_prompt(metrics, typical, word_limit=word_limit),
+            system_prompt="只返回一段中文总结，不使用 Markdown。",
+        )
+
+    return summarize_report
+
+
+def build_ai_dependencies():
+    client = build_ai_client()
+    requester = build_ai_requester(client)
+    return requester, build_report_summary_provider(client)
+
+
 def create_task(session: Session, request: CreateTaskRequest, *, submit_background: bool = True) -> Task:
     task = Task(
         platform=request.platform,
@@ -108,6 +136,8 @@ def run_task(task_id: str, *, session_factory: Callable[[], Session] | sessionma
             return
         try:
             platform = build_platform(task.platform)
+            ai_requester, report_summary_provider = build_ai_dependencies()
+
             try:
                 hotspots = platform.fetch_hotspots(limit=task.hotspot_limit)
             except PlatformAPIError as exc:
@@ -183,7 +213,7 @@ def run_task(task_id: str, *, session_factory: Callable[[], Session] | sessionma
                         item_comments = list(session.scalars(select(Comment).where(Comment.content_item_id == item.id)))
                         ai_results = analyze_comments_with_retry(
                             [{"comment_id": comment.id, "content": comment.content} for comment in item_comments],
-                            requester=build_ai_requester(),
+                            requester=ai_requester,
                             max_retries=get_settings().ai_max_retries,
                         )
                         results_by_comment_id = {result.comment_id: result for result in ai_results}
@@ -201,7 +231,7 @@ def run_task(task_id: str, *, session_factory: Callable[[], Session] | sessionma
                             comment.ai_analysis_status = result.ai_analysis_status
                         item.status = "success"
                         task.successful_items_count += 1
-                        generate_item_report(session, item.id)
+                        generate_item_report(session, item.id, summary_provider=report_summary_provider)
                     except Exception as exc:
                         item.status = "failed"
                         item.error_stage = "crawl_comments"
@@ -223,7 +253,7 @@ def run_task(task_id: str, *, session_factory: Callable[[], Session] | sessionma
                     total_count=total_comments,
                 )
                 for hotspot in task.hotspots:
-                    generate_hotspot_report(session, hotspot.id)
+                    generate_hotspot_report(session, hotspot.id, summary_provider=report_summary_provider)
                 task.status = "success"
             else:
                 _mark_task_failed(task, task.error_stage or "crawl_items", task.error_type or "no_successful_items", task.error_message or "没有任何内容条目成功")

docker-compose.yml

@@ -1,3 +1,5 @@
+name: hot-comments-tool
+
 services:
   app:
     build: .

tests/integration/test_task_flow_xiaohongshu.py

@@ -5,6 +5,7 @@ from sqlalchemy.orm import Session
 
 from app.models import Comment, ContentItem, Hotspot, Report, Task
 from app.schemas import CreateTaskRequest
+from app.services.report_service import DEFAULT_SUMMARY
 from app.services.task_service import create_task, run_task
 from tests.helpers import make_test_client
 
@@ -28,7 +29,11 @@ class FakeXhsPlatform:
         return [CommentData(source_comment_id="c1", content="评论一", like_count=5, raw_data={"id": "c1"})]
 
 
+ai_prompts: list[str] = []
+
+
 def successful_ai_response(prompt: str) -> str:
+    ai_prompts.append(prompt)
     comments = json.loads(prompt[prompt.index("[") :])
     comment_id = comments[0]["comment_id"]
     return json.dumps(
@@ -38,9 +43,14 @@ def successful_ai_response(prompt: str) -> str:
 
 
 def test_xiaohongshu_task_flow_persists_hotspots_items_and_comments(monkeypatch):
+    ai_prompts.clear()
     with make_test_client() as (_client, engine):
         monkeypatch.setattr("app.services.task_service.build_platform", lambda _platform: FakeXhsPlatform())
-        monkeypatch.setattr("app.services.task_service.build_ai_requester", lambda: successful_ai_response)
+        summary_provider = lambda metrics, typical, *, word_limit=200: (
+            ai_prompts.append(f"只返回一段中文总结 word_limit={word_limit} sample={metrics['sample_count']}")
+            or "这是一段真实 AI 报告摘要。"
+        )
+        monkeypatch.setattr("app.services.task_service.build_ai_dependencies", lambda: (successful_ai_response, summary_provider))
 
         with Session(engine) as session:
             task = create_task(
@@ -66,14 +76,21 @@ def test_xiaohongshu_task_flow_persists_hotspots_items_and_comments(monkeypatch)
             assert comment.sentiment == "positive"
             assert comment.labels == '["认可"]'
             assert comment.reason == "喜欢"
-            assert session.scalar(select(Report).where(Report.task_id == task_id, Report.report_type == "item")) is not None
+            item_report = session.scalar(select(Report).where(Report.task_id == task_id, Report.report_type == "item"))
-            assert session.scalar(select(Report).where(Report.task_id == task_id, Report.report_type == "hotspot")) is not None
+            hotspot_report = session.scalar(select(Report).where(Report.task_id == task_id, Report.report_type == "hotspot"))
+            assert item_report is not None
+            assert hotspot_report is not None
+            assert item_report.summary == "这是一段真实 AI 报告摘要。"
+            assert hotspot_report.summary == "这是一段真实 AI 报告摘要。"
+            assert DEFAULT_SUMMARY not in item_report.markdown_content
+            assert DEFAULT_SUMMARY not in hotspot_report.markdown_content
+            assert any("只返回一段中文总结" in prompt for prompt in ai_prompts)
 
 
 def test_xiaohongshu_task_flow_marks_comments_failed_when_ai_parse_fails(monkeypatch):
     with make_test_client() as (_client, engine):
         monkeypatch.setattr("app.services.task_service.build_platform", lambda _platform: FakeXhsPlatform())
-        monkeypatch.setattr("app.services.task_service.build_ai_requester", lambda: (lambda _prompt: "not-json"))
+        monkeypatch.setattr("app.services.task_service.build_ai_dependencies", lambda: (lambda _prompt: "not-json", None))
         monkeypatch.setattr("app.services.ai_service.time.sleep", lambda _seconds: None)
 
         with Session(engine) as session:
@@ -107,7 +124,7 @@ def test_xiaohongshu_task_flow_keeps_item_success_when_ai_request_raises(monkeyp
         def failing_requester(_prompt):
             raise RuntimeError("ai unauthorized")
 
-        monkeypatch.setattr("app.services.task_service.build_ai_requester", lambda: failing_requester)
+        monkeypatch.setattr("app.services.task_service.build_ai_dependencies", lambda: (failing_requester, None))
 
         with Session(engine) as session:
             task = create_task(

tests/unit/test_ai_schema.py

@@ -5,10 +5,12 @@ from app.services.ai_service import (
     AIAnalysisResult,
     OpenAICompatibleAIClient,
     analyze_comments_with_retry,
+    build_report_summary_prompt,
     build_comment_prompt,
     calculate_analysis_status,
     parse_ai_comment_response,
 )
+from app.services.task_service import build_report_summary_provider
 
 
 def test_build_comment_prompt_contains_ids_and_truncates_content():
@@ -20,6 +22,35 @@ def test_build_comment_prompt_contains_ids_and_truncates_content():
     assert "你" * 151 not in prompt
 
 
+def test_build_report_summary_prompt_contains_stats_labels_and_truncated_typical_comments():
+    prompt = build_report_summary_prompt(
+        {
+            "sample_count": 2,
+            "sentiment": {
+                "positive": {"count": 1, "pct": 50.0},
+                "neutral": {"count": 0, "pct": 0.0},
+                "negative": {"count": 1, "pct": 50.0},
+                "unknown": {"count": 0, "pct": 0.0},
+            },
+            "top_labels": [{"name": "价格争议", "count": 2}],
+        },
+        {"positive": [{"content": "好" * 200}], "negative": [{"content": "太贵"}]},
+        word_limit=200,
+    )
+
+    assert "只返回一段中文总结" in prompt
+    assert "样本评论数量：2" in prompt
+    assert "positive: 1 (50.0%)" in prompt
+    assert "价格争议: 2" in prompt
+    assert "好" * 150 in prompt
+    assert "好" * 151 not in prompt
+    assert "JSON Array" not in prompt
+
+
+def test_report_summary_provider_is_disabled_without_real_ai_client():
+    assert build_report_summary_provider(client=None) is None
+
+
 def test_parse_ai_comment_response_validates_array_sentiment_labels_and_ids():
     result = parse_ai_comment_response(
         '[{"comment_id":"c1","sentiment":"positive","labels":["质量好"],"reason":"认可"}]',
@@ -94,6 +125,27 @@ def test_openai_compatible_client_posts_chat_completion_and_returns_message_cont
     assert captured["authorization"] == "Bearer test-ai-key"
     assert '"model":"test-model"' in captured["body"]
     assert "prompt text" in captured["body"]
+    assert "请严格遵循用户指令输出。" in captured["body"]
+
+
+def test_openai_compatible_client_accepts_custom_system_prompt():
+    captured = {}
+
+    def handler(request: httpx.Request) -> httpx.Response:
+        captured["body"] = request.read().decode("utf-8")
+        return httpx.Response(200, json={"choices": [{"message": {"content": "报告摘要"}}]})
+
+    http_client = httpx.Client(transport=httpx.MockTransport(handler))
+    client = OpenAICompatibleAIClient(
+        base_url="https://ai.example.com",
+        api_key="test-ai-key",
+        model="test-model",
+        http_client=http_client,
+    )
+
+    assert client.request("总结 prompt", system_prompt="只返回一段中文总结，不使用 Markdown。") == "报告摘要"
+    assert "只返回一段中文总结" in captured["body"]
+    assert "JSON Array" not in captured["body"]
 
 
 def test_openai_compatible_client_accepts_base_url_that_already_includes_v1():

tests/unit/test_deployment_config.py

@@ -11,6 +11,12 @@ def test_docker_compose_reads_real_env_file_not_example():
     assert "- .env.example" not in compose_content
 
 
+def test_docker_compose_uses_fixed_project_name_for_stable_port_owner():
+    compose_content = (ROOT / "docker-compose.yml").read_text(encoding="utf-8")
+
+    assert "name: hot-comments-tool" in compose_content
+
+
 def test_real_env_file_is_gitignored():
     gitignore_lines = (ROOT / ".gitignore").read_text(encoding="utf-8").splitlines()
 

tests/unit/test_douyin_mapping.py

@@ -118,3 +118,39 @@ def test_douyin_maps_null_comments_as_empty_list():
     payload = {"data": {"comments": None}}
 
     assert platform.map_comments(payload, limit=10) == []
+
+
+class RecordingDouyinClient:
+    def __init__(self):
+        self.requests = []
+
+    def get(self, path, *, params=None):
+        self.requests.append((path, params))
+        cursor = params["cursor"]
+        if cursor == 0:
+            return {
+                "data": {
+                    "comments": [{"cid": f"c-{index}", "text": f"评论 {index}"} for index in range(20)],
+                    "cursor": 20,
+                    "has_more": 1,
+                }
+            }
+        return {
+            "data": {
+                "comments": [{"cid": f"c-{index}", "text": f"评论 {index}"} for index in range(20, 55)],
+                "cursor": 55,
+                "has_more": 0,
+            }
+        }
+
+
+def test_douyin_fetch_comments_paginates_until_limit():
+    client = RecordingDouyinClient()
+    platform = DouyinPlatform(client=client)
+
+    comments = platform.fetch_comments("aweme-1", limit=50)
+
+    assert len(comments) == 50
+    assert comments[0].source_comment_id == "c-0"
+    assert comments[-1].source_comment_id == "c-49"
+    assert [request[1]["cursor"] for request in client.requests] == [0, 20]

tests/unit/test_task_executor.py

@@ -1,6 +1,7 @@
 from concurrent.futures import ThreadPoolExecutor
 
 from app.services import task_service
+from app.config import Settings
 
 
 def test_task_executor_is_single_worker_thread_pool():
@@ -10,3 +11,12 @@ def test_task_executor_is_single_worker_thread_pool():
 
 def test_run_task_entrypoint_is_synchronous_function():
     assert task_service.inspect.iscoroutinefunction(task_service.run_task) is False
+
+
+def test_ai_dependencies_do_not_generate_report_summary_provider_without_ai_config(monkeypatch):
+    monkeypatch.setattr("app.services.task_service.get_settings", lambda: Settings(_env_file=None))
+
+    requester, summary_provider = task_service.build_ai_dependencies()
+
+    assert requester("prompt") == "[]"
+    assert summary_provider is None