|
|
@@ -278,54 +278,7 @@
|
|
|
")\n",
|
|
|
"\n",
|
|
|
"# ==============================================================================\n",
|
|
|
- "# 2. 提取专家 (Extractor) - 数据清洗 (加入反思机制)\n",
|
|
|
- "# ==============================================================================\n",
|
|
|
- "EXTRACTOR_PROMPTS = {\n",
|
|
|
- " \"initial\": \"\"\"\n",
|
|
|
- " 你是一个 **纯粹的数据提取器**。\n",
|
|
|
- " 你的任务是从 `<<<原始数据>>>` 中提取商品参数。\n",
|
|
|
- "\n",
|
|
|
- " ### 🧠 CoT 提取逻辑\n",
|
|
|
- " 1. **扫描**: 快速浏览文本，寻找 \"¥\", \"$\", \"http\" 等关键符号。\n",
|
|
|
- " 2. **验证**: 确认找到的信息是否属于\"核心商品\"（而非广告推荐）。\n",
|
|
|
- " 3. **格式化**: 将提取到的信息填入 JSON。\n",
|
|
|
- "\n",
|
|
|
- " ### 🚫 绝对禁令\n",
|
|
|
- " 1. **严禁伪造**: 没找到价格就填 \"暂无报价\"，绝对不要编数字！\n",
|
|
|
- " 2. **严禁 API 格式**: 不要输出 `{{ \"status\": ... }}`。\n",
|
|
|
- " 3. **严禁 Markdown**: 直接输出列表 `[...]`。\n",
|
|
|
- "\n",
|
|
|
- " ### ✅ 正确范例\n",
|
|
|
- " 输入: ...Dior 口红...价格 ¥350...图片 http://img...\n",
|
|
|
- " 输出:\n",
|
|
|
- " [\n",
|
|
|
- " {{\n",
|
|
|
- " \"name\": \"Dior 口红\",\n",
|
|
|
- " \"price\": \"¥350\",\n",
|
|
|
- " \"img\": \"http://img...\"\n",
|
|
|
- " }}\n",
|
|
|
- " ]\n",
|
|
|
- "\n",
|
|
|
- " ---\n",
|
|
|
- " <<<原始数据开始>>>\n",
|
|
|
- " {task}\n",
|
|
|
- " <<<原始数据结束>>>\n",
|
|
|
- " \n",
|
|
|
- " 请提取数据并输出 JSON 列表：\n",
|
|
|
- " \"\"\",\n",
|
|
|
- " \"reflect\": \"检查输出：是否以 `[` 开头？是否包含幻觉商品？\",\n",
|
|
|
- " \"refine\": \"修正格式，只输出纯净的 JSON 列表 `[...]`。\"\n",
|
|
|
- "}\n",
|
|
|
- "\n",
|
|
|
- "extractor_agent = ReflectionAgent(\n",
|
|
|
- " name=\"Agent_Extractor\",\n",
|
|
|
- " llm=llm,\n",
|
|
|
- " max_iterations=1, \n",
|
|
|
- " custom_prompts=EXTRACTOR_PROMPTS\n",
|
|
|
- ")\n",
|
|
|
- "\n",
|
|
|
- "# ==============================================================================\n",
|
|
|
- "# 3. 种草达人 (Pitcher) - 文案创作 (加入风格指导)\n",
|
|
|
+ "# 2. 种草达人 (Pitcher) - 文案创作 (加入风格指导)\n",
|
|
|
"# ==============================================================================\n",
|
|
|
"PITCHER_PROMPT = \"\"\"\n",
|
|
|
"你是一个 **金牌种草文案**。\n",
|
jack6249