fix: AI分析报告无响应 — 修复系统提示词和工具描述，引导模型主动调用工具查询数据

app/routes/ai_routes.py

@@ -18,7 +18,7 @@ TOOLS = [
         "type": "function",
         "function": {
             "name": "query_database",
-            "description": "Execute a SQL SELECT query to retrieve data. Available tables:\n1. deep_collection (id, url, content, summary, status, error_msg, created_at, updated_at)\n2. spider_result (id, task_id, title, abstract, source, cover, link, created_at)\n3. collection_task (id, keyword, source, status, created_at, finished_at)\nUse SQLite syntax.",
+            "description": "Execute a SQL SELECT query to retrieve data from a PostgreSQL database. Available tables:\n1. deep_collection (id, url, content, summary, status, error_msg, created_at, updated_at)\n2. spider_result (id, task_id, title, abstract, source, cover, link, created_at)\n3. collection_task (id, keyword, source, status, created_at, finished_at)\nUse PostgreSQL syntax. Use COUNT/GROUP BY for statistics. Always call this tool when the user asks about data.",
             "parameters": {
                 "type": "object",
                 "properties": {
@@ -384,11 +384,17 @@ def analysis_chat():
     messages = []
     system_prompt = model.system_prompt if model.system_prompt else "You are a helpful AI assistant."
     system_prompt += """
-    \nIMPORTANT:
-    1. You MUST use the `render_chart` tool to display charts. NEVER generate Python code (like matplotlib/pyecharts) for charts in the text response.
-    2. When using `render_chart`, you can generate "combined" charts by setting `type="combined"` and providing data for "bar", "pie", etc.
-    3. Use Markdown tables for data presentation when not using charts.
-    """
+
+你是一个专业的数据分析助手。你可以访问以下工具：
+1. query_database - 查询 PostgreSQL 数据库中的采集数据。当用户询问任何关于数据的问题时，你**必须首先调用此工具**查询数据，不要凭空回答。
+2. render_chart - 生成图表展示数据。查询到数据后，使用此工具生成可视化图表。
+
+工作流程：
+- 收到用户问题 → 调用 query_database 查询相关数据 → 根据查询结果调用 render_chart 生成图表 → 用文字总结分析结果
+- SQL 使用 PostgreSQL 语法
+- 统计类问题使用 COUNT + GROUP BY
+- 回答使用中文
+"""
     messages.append({"role": "system", "content": system_prompt})
     
     for m in past_messages: