add prompt assistant

prompt/consumers.py

@@ -3,7 +3,7 @@ from channels.generic.websocket import AsyncWebsocketConsumer
 from channels.db import database_sync_to_async
 from django.db.models import Count
 from .models import Conversation, Message
-from .llm import stream_chat, extract_code
+from .llm import stream_chat, extract_code, stream_guidance, parse_guidance_response
 
 
 class PromptConsumer(AsyncWebsocketConsumer):
@@ -126,3 +126,93 @@ class PromptConsumer(AsyncWebsocketConsumer):
     def get_history_for_llm(self):
         messages = self.conversation.messages.filter(source="conversation")
         return [{"role": m.role, "content": m.content} for m in messages]
+
+
+class GuidanceConsumer(AsyncWebsocketConsumer):
+    async def connect(self):
+        self.user = self.scope["user"]
+        if self.user.is_anonymous:
+            await self.close()
+            return
+
+        self.task_id = int(self.scope["url_route"]["kwargs"]["task_id"])
+        self.current_user_message = None
+        self.session_messages = []
+        await self.accept()
+
+        self.conversation = await self.get_or_create_conversation()
+        await self.send(text_data=json.dumps({"type": "init", "messages": []}))
+
+    async def disconnect(self, close_code):
+        if self.current_user_message:
+            await self.delete_message(self.current_user_message)
+            self.current_user_message = None
+
+    async def receive(self, text_data):
+        data = json.loads(text_data)
+        prompt = data.get("content", "").strip()
+        if not prompt:
+            return
+
+        self.current_user_message = await self.save_message("user", prompt)
+        self.session_messages.append({"role": "user", "content": prompt})
+
+        try:
+            full_response = ""
+            try:
+                async for chunk in stream_guidance(self.session_messages):
+                    full_response += chunk
+                    await self.send(text_data=json.dumps({
+                        "type": "stream",
+                        "content": chunk,
+                    }))
+            except Exception as e:
+                await self.send(text_data=json.dumps({
+                    "type": "error",
+                    "content": f"AI 服务出错：{str(e)}",
+                }))
+                return
+
+            clean_response, is_ready = parse_guidance_response(full_response)
+            self.session_messages.append({
+                "role": "assistant",
+                "content": clean_response,
+            })
+            assistant_msg = await self.save_message("assistant", clean_response)
+            self.current_user_message = None
+
+            await self.send(text_data=json.dumps({
+                "type": "complete",
+                "message_id": assistant_msg.id,
+                "is_ready": is_ready,
+            }))
+
+        finally:
+            if self.current_user_message:
+                await self.delete_message(self.current_user_message)
+                self.current_user_message = None
+
+    @database_sync_to_async
+    def get_or_create_conversation(self):
+        conv = (
+            Conversation.objects.filter(user=self.user, task_id=self.task_id)
+            .annotate(msg_count=Count("messages"))
+            .order_by("-msg_count", "-created")
+            .first()
+        )
+        if not conv:
+            conv = Conversation.objects.create(user=self.user, task_id=self.task_id)
+        return conv
+
+    @database_sync_to_async
+    def delete_message(self, message):
+        message.delete()
+
+    @database_sync_to_async
+    def save_message(self, role, content):
+        return Message.objects.create(
+            conversation=self.conversation,
+            role=role,
+            source="guidance",
+            content=content,
+        )

prompt/llm.py

@@ -45,9 +45,11 @@ GUIDANCE_SYSTEM_PROMPT = """你是一个提示词写作教练，帮助学生写
 1. 如果提示词不够好，用 1-2 个启发性问题引导学生补充细节，不要直接给出答案
 2. 如果提示词已经够好，以 [READY] 开头回复，简短夸奖学生并说明可以生成了
 3. 用中文回复，语气鼓励，简洁明了
-4. 不要生成任何代码"""
+4. 使用 Markdown 语法高亮关键词，优先突出 **主题**、**视觉**、**交互**、**内容**、**可以生成** 等重点
+5. 如果回复以 [READY] 开头，[READY] 不要加粗，必须保持原始文本
+6. 不要生成任何代码"""
 
-DEFAULT_MODEL = "deepseek-chat"
+DEFAULT_MODEL = "deepseek-v4-flash"
 
 # Models served by the ARK (Volcengine) endpoint
 ARK_MODELS = {"doubao-seed-2-0-lite-260215"}

prompt/tests.py

@@ -297,10 +297,16 @@ class PromptHistoryTest(TestCase):
         self.assertIsNone(data[0]["code_js"])
 
 
-from prompt.llm import parse_guidance_response
+from prompt.llm import GUIDANCE_SYSTEM_PROMPT, parse_guidance_response
 
 
 class ParseGuidanceResponseTest(TestCase):
+    def test_guidance_prompt_asks_for_bold_keywords(self):
+        self.assertIn("Markdown", GUIDANCE_SYSTEM_PROMPT)
+        self.assertIn("[READY] 不要加粗", GUIDANCE_SYSTEM_PROMPT)
+        for keyword in ("**主题**", "**视觉**", "**交互**", "**内容**", "**可以生成**"):
+            self.assertIn(keyword, GUIDANCE_SYSTEM_PROMPT)
+
     def test_ready_prefix_with_newline_stripped(self):
         content, is_ready = parse_guidance_response("[READY]\n很好，可以生成了！")
         self.assertEqual(content, "很好，可以生成了！")

prompt/url.py

@@ -1,6 +1,15 @@
-from django.urls import path
-from .consumers import PromptConsumer
+from collections.abc import Callable
+from typing import Any, cast
 
-websocket_urlpatterns = [
-    path("ws/prompt/<int:task_id>/", PromptConsumer.as_asgi()),
+from django.urls import path
+from django.urls.resolvers import URLPattern, URLResolver
+
+from .consumers import PromptConsumer, GuidanceConsumer
+
+AsgiApplication = Callable[..., Any]
+RoutePattern = URLPattern | URLResolver
+
+websocket_urlpatterns: list[RoutePattern] = [
+    path("ws/prompt/<int:task_id>/", cast(AsgiApplication, PromptConsumer.as_asgi())),
+    path("ws/guidance/<int:task_id>/", cast(AsgiApplication, GuidanceConsumer.as_asgi())),
 ]