\u524d\u9762\u5df2\u7ecf\u642d\u597d\u4e86：<\/p>\n
\u4f46\u5728\u771f\u5b9e\u4f7f\u7528\u4e2d，\u4f60\u4e00\u5b9a\u5df2\u7ecf\u9047\u5230\u8fc7\u8fd9\u79cd\u60c5\u51b5：<\/p>\n
\u6839\u672c\u539f\u56e0\u662f：<\/p>\n
\u6211\u4eec\u73b0\u5728\u7684\u5927\u90e8\u5206 Agent，\u53ea\u6709\u201c\u6b63\u5411\u751f\u6210\u201d\u7684\u672c\u4e8b，
\n\u6ca1\u6709\u201c\u53cd\u5411\u68c0\u67e5\u201d\u548c\u201c\u81ea\u6211\u6000\u7591\u201d\u7684\u673a\u5236\u3002<\/p>\n
\u4e13\u95e8\u505a\u4e00\u4ef6\u4e8b：<\/p>\n
\u5728\u73b0\u6709\u7684 Agent \u4f53\u7cfb\u4e4b\u4e0a，\u52a0\u4e00\u5c42\u81ea\u6211\u53cd\u601d（Self-Reflection）\u80fd\u529b，
\n\u8ba9 Agent \u5728\u7ed9\u7528\u6237\u7b54\u6848\u4e4b\u524d，\u5148\u5ba1\u67e5\u4e00\u4e0b\u81ea\u5df1\u6709\u6ca1\u6709\u80e1\u8bf4\u3002<\/p>\n
\u8fd9\u91cc\u7684\u201c\u81ea\u6211\u53cd\u601d\u201d，\u4e0d\u662f\u54f2\u5b66\u4e0a\u7684\u90a3\u4e00\u5957，\u800c\u662f\u5f88\u5de5\u7a0b\u5316\u7684\u4e1c\u897f，\u53ef\u4ee5\u7406\u89e3\u4e3a：<\/p>\n
\u5728\u8f93\u51fa\u4e4b\u524d，\u5f3a\u5236\u518d\u8dd1\u4e00\u904d**\u201c\u6162\u601d\u8003（\u5ba1\u7a3f\u4eba\u6a21\u5f0f）\u201d**，
\n\u5bf9\u521a\u624d\u7684\u7b54\u6848\u505a\u4e00\u6b21\u7ed3\u6784\u5316\u7684\u68c0\u67e5\u548c\u4fee\u8ba2\u3002<\/p>\n
\u5b83\u81f3\u5c11\u5305\u62ec\u4e09\u7c7b\u68c0\u67e5：<\/p>\n
\u4e8b\u5b9e\u68c0\u67e5<\/p>\n
\u903b\u8f91\u68c0\u67e5<\/p>\n
\u8bed\u6c14\u4e0e\u7f6e\u4fe1\u5ea6\u68c0\u67e5<\/p>\n
\u5927\u81f4\u6d41\u7a0b\u53ef\u4ee5\u60f3\u8c61\u6210：<\/p>\n
\u7528\u6237\u95ee\u9898 \u2192（\u5feb\u601d\u8003）\u751f\u6210\u521d\u7a3f \u2192（\u6162\u601d\u8003）\u81ea\u6211\u5ba1\u67e5 \u2192 \u4fee\u8ba2\u7248 \u2192 \u8fd4\u56de\u7528\u6237<\/p>\n
\u6839\u636e\u4f60\u73b0\u5728\u7684\u7cfb\u7edf\u5f62\u6001\u548c\u9884\u7b97，\u53ef\u4ee5\u8003\u8651\u4e24\u79cd\u4e3b\u8981\u6a21\u5f0f：<\/p>\n
\u4e00\u4e2a Agent \u5185\u90e8\u5305\u542b\u4e24\u4e2a\u9636\u6bb5：<\/p>\n
\u7279\u70b9：<\/p>\n
\u628a\u89d2\u8272\u62c6\u5f00：<\/p>\n
\u7279\u70b9：<\/p>\n
\u4e0b\u9762\u6211\u7528\u53cc Agent \u6a21\u5f0f\u4f5c\u4e3a\u4e3b\u7ebf\u6765\u8bb2\u89e3，\u6700\u540e\u518d\u7ed9\u4e00\u4e2a\u5355 Agent \u7248\u672c\u7684\u8f7b\u91cf\u5b9e\u73b0\u3002<\/p>\n
\u6211\u4eec\u5148\u62bd\u8c61\u4e00\u4e2a\u201c\u5e26\u81ea\u6211\u53cd\u601d\u80fd\u529b\u7684 Agent\u201d：<\/p>\n
from typing import Dict, Any, List
\nimport json<\/p>\n
class SelfReflectingAgent:
\n """
\n \u53ccAgent\u67b6\u6784：
\n – generator_llm: \u8d1f\u8d23\u751f\u6210\u7b54\u6848
\n – reviewer_llm: \u8d1f\u8d23\u5ba1\u67e5\u7b54\u6848
\n """
\n def __init__(self, generator_llm, reviewer_llm, max_reflection: int = 2):
\n self.generator = generator_llm
\n self.reviewer = reviewer_llm
\n self.max_reflection = max_reflection<\/p>\n
# — \u5feb\u601d\u8003：\u751f\u6210\u521d\u7a3f —
\n def generate_answer(self, question: str) -> str:
\n prompt = f"\u8bf7\u76f4\u63a5\u3001\u7b80\u660e\u5730\u56de\u7b54\u8fd9\u4e2a\u95ee\u9898：\\\\n{question}"
\n return self.generator(prompt)<\/p>\n
# — \u6162\u601d\u8003：\u7ed3\u6784\u5316\u5ba1\u67e5 —
\n def review_answer(self, question: str, answer: str, docs: List[str]) -> Dict[str, Any]:
\n review_prompt = self._build_review_prompt(question, answer, docs)
\n resp = self.reviewer(review_prompt)
\n try:
\n data = json.loads(resp)
\n except json.JSONDecodeError:
\n data = {
\n "valid": False,
\n "issues": [{"type": "\u683c\u5f0f\u9519\u8bef", "description": "\u5ba1\u6838\u8f93\u51fa\u975eJSON", "location": ""}],
\n "corrections": [],
\n "confidence": 0.0
\n }
\n return data<\/p>\n
# — \u5e94\u7528\u5ba1\u7a3f\u5efa\u8bae：\u7b80\u5355\u7248\u672c\u7528\u66ff\u6362 —
\n def apply_corrections(self, answer: str, corrections: List[Dict[str, str]]) -> str:
\n for c in corrections:
\n answer = answer.replace(c.get("original", ""), c.get("corrected", ""))
\n return answer<\/p>\n
# — \u6574\u4f53\u5165\u53e3 —
\n def run(self, question: str, docs: List[str]) -> Dict[str, Any]:
\n """
\n \u8fd4\u56de\u7ed3\u6784：
\n {
\n "final_answer": str,
\n "review": {…},
\n "iterations": int,
\n "status": "success|failure"
\n }
\n """
\n answer = self.generate_answer(question)<\/p>\n
review = {}
\n for i in range(self.max_reflection):
\n review = self.review_answer(question, answer, docs)<\/p>\n
# \u6ee1\u8db3\u201c\u5ba1\u6838\u901a\u8fc7\u201d\u7684\u6761\u4ef6\u76f4\u63a5\u8fd4\u56de
\n if review.get("valid") and review.get("confidence", 0.0) >= 0.9:
\n return {
\n "final_answer": answer,
\n "review": review,
\n "iterations": i + 1,
\n "status": "success"
\n }<\/p>\n
# \u6709\u660e\u786e\u7684 corrections，\u5c1d\u8bd5\u81ea\u52a8\u4fee\u6b63
\n if review.get("corrections"):
\n answer = self.apply_corrections(answer, review["corrections"])
\n else:
\n # \u6ca1\u6cd5\u4fee，\u8bf4\u660e\u95ee\u9898\u6bd4\u8f83\u5927
\n return {
\n "final_answer": "\u5f53\u524d\u65e0\u6cd5\u751f\u6210\u53ef\u9760\u7b54\u6848，\u8bf7\u7a0d\u540e\u518d\u8bd5，\u6216\u4ea4\u7531\u4eba\u5de5\u5904\u7406\u3002",
\n "review": review,
\n "iterations": i + 1,
\n "status": "failure"
\n }<\/p>\n
# \u8fbe\u5230\u6700\u5927\u53cd\u601d\u8f6e\u6570\u4ecd\u7136\u4e0d\u8fc7\u5ba1
\n return {
\n "final_answer": "\u7ecf\u8fc7\u591a\u8f6e\u68c0\u67e5\u4ecd\u65e0\u6cd5\u786e\u8ba4\u7b54\u6848\u53ef\u9760，\u5efa\u8bae\u4eba\u5de5\u590d\u6838\u3002",
\n "review": review,
\n "iterations": self.max_reflection,
\n "status": "failure"
\n }<\/p>\n
\u81ea\u6211\u53cd\u601d\u80fd\u4e0d\u80fd\u843d\u5730，\u5f88\u5927\u7a0b\u5ea6\u4e0a\u53d6\u51b3\u4e8e\u5ba1\u6838 Prompt \u5199\u5f97\u597d\u4e0d\u597d\u3002<\/p>\n
\u5bf9\u5ba1\u6838 Agent \u6765\u8bf4，\u4f60\u8981\u544a\u8bc9\u5b83：<\/p>\n
\u4e00\u4e2a\u5b9e\u7528\u7684\u5b9e\u73b0：<\/p>\n
class ReviewPromptBuilder:
\n def __init__(self, max_docs: int = 5):
\n self.max_docs = max_docs<\/p>\n
def build(self, question: str, answer: str, docs: List[str]) -> str:
\n doc_block = "\\\\n".join(
\n [f"[\u6587\u6863{i+1}]\\\\n{d[:300]}…" for i, d in enumerate(docs[:self.max_docs])]
\n )<\/p>\n
return f"""
\n\u4f60\u662f\u4e00\u4e2a\u4e25\u683c\u7684\u5ba1\u7a3f\u4eba，\u9700\u8981\u6839\u636e\u7ed9\u5b9a\u7684\u6587\u6863\u5ba1\u67e5\u4e00\u4e2a\u7b54\u6848\u662f\u5426\u53ef\u9760\u3002<\/p>\n
# \u95ee\u9898
\n{question}<\/p>\n
# \u53c2\u8003\u6587\u6863（\u622a\u65ad\u663e\u793a）
\n{doc_block}<\/p>\n
# \u5f85\u5ba1\u6838\u7b54\u6848
\n{answer}<\/p>\n
# \u5ba1\u67e5\u4efb\u52a1（\u8bf7\u9010\u6761\u68c0\u67e5）
\n1. \u6587\u6863\u5f15\u7528\u662f\u5426\u51c6\u786e？
\n – \u6bcf\u6761\u4e8b\u5b9e\u6027\u9648\u8ff0\u662f\u5426\u80fd\u5728\u4e0a\u8ff0\u6587\u6863\u4e2d\u627e\u5230\u652f\u6301？
\n – \u662f\u5426\u5f15\u7528\u4e86\u6587\u6863\u4e2d\u4e0d\u5b58\u5728\u7684\u5185\u5bb9？<\/p>\n
2. \u662f\u5426\u5b58\u5728\u4e8b\u5b9e\u6027\u9519\u8bef？
\n – \u6570\u5b57\u3001\u65e5\u671f\u3001\u540d\u79f0\u662f\u5426\u4e0e\u6587\u6863\u4e00\u81f4？
\n – \u63a8\u8bba\u662f\u5426\u4e0e\u6587\u6863\u63cf\u8ff0\u77db\u76fe？<\/p>\n
3. \u903b\u8f91\u662f\u5426\u81ea\u6d3d？
\n – \u540c\u4e00\u7b54\u6848\u5185\u90e8\u6709\u65e0\u77db\u76fe？
\n – \u662f\u5426\u6709\u660e\u663e\u7684\u903b\u8f91\u6f0f\u6d1e？<\/p>\n
4. \u8bed\u6c14\u662f\u5426\u5408\u7406？
\n – \u662f\u5426\u6709\u201c\u80af\u5b9a\/\u5fc5\u987b\/\u7edd\u5bf9\u201d\u7b49\u8fc7\u5ea6\u81ea\u4fe1\u7684\u8868\u8fbe，\u5728\u6ca1\u6709\u6587\u6863\u652f\u6301\u7684\u60c5\u51b5\u4e0b\u4f7f\u7528？<\/p>\n
# \u8f93\u51fa\u8981\u6c42
\n\u4ec5\u8f93\u51fa\u4e00\u6bb5 JSON \u6587\u672c，\u683c\u5f0f\u5982\u4e0b：
\n{{
\n "valid": true\/false,
\n "issues": [
\n {{
\n "type": "\u5f15\u7528\u9519\u8bef|\u4e8b\u5b9e\u9519\u8bef|\u903b\u8f91\u9519\u8bef|\u8fc7\u5ea6\u81ea\u4fe1",
\n "description": "\u95ee\u9898\u63cf\u8ff0",
\n "location": "\u76f8\u5173\u53e5\u5b50\u6216\u7247\u6bb5"
\n }}
\n ],
\n "corrections": [
\n {{
\n "type": "\u5f15\u7528\u4fee\u6b63|\u4e8b\u5b9e\u4fee\u6b63|\u903b\u8f91\u4fee\u6b63|\u8bed\u6c14\u4fee\u6b63",
\n "original": "\u539f\u53e5\u5b50",
\n "corrected": "\u4fee\u6b63\u540e\u7684\u53e5\u5b50"
\n }}
\n ],
\n "confidence": 0.0-1.0,
\n "reasoning": "\u7b80\u8981\u8bf4\u660e\u4f60\u505a\u51fa\u4e0a\u8ff0\u5224\u65ad\u7684\u539f\u56e0"
\n}}<\/p>\n
\u4e0d\u8981\u8f93\u51fa\u4efb\u4f55\u591a\u4f59\u89e3\u91ca\u3002
\n"""<\/p>\n
\u7136\u540e\u5728 SelfReflectingAgent \u4e2d\u7528\u5b83：<\/p>\n
class SelfReflectingAgent:
\n def __init__(…, …):
\n …
\n self.prompt_builder = ReviewPromptBuilder()<\/p>\n
def _build_review_prompt(self, question, answer, docs):
\n return self.prompt_builder.build(question, answer, docs)<\/p>\n
\u5982\u679c\u4f60\u6682\u65f6\u4e0d\u60f3\u4e0a\u201c\u53cc Agent \u6a21\u5f0f\u201d，\u6216\u8005\u53ea\u60f3\u5728\u90e8\u5206\u573a\u666f\u91cc\u505a\u4e2a\u8bd5\u70b9，\u53ef\u4ee5\u7528\u5355 Agent + \u53cc\u9636\u6bb5 Prompt\u7684\u505a\u6cd5\u3002<\/p>\n
def self_reflecting_answer(llm, question: str, docs: List[str]) -> str:
\n # \u7b2c\u4e00\u8f6e：\u6b63\u5e38\u56de\u7b54
\n context = "\\\\n\\\\n".join(docs[:3])
\n answer_prompt = f"""
\n\u4f60\u662f\u4e00\u4e2a\u95ee\u7b54\u52a9\u624b，\u8bf7\u57fa\u4e8e\u4ee5\u4e0b\u4e0a\u4e0b\u6587\u56de\u7b54\u95ee\u9898：<\/p>\n
\u4e0a\u4e0b\u6587：
\n{context}<\/p>\n
\u95ee\u9898：
\n{question}<\/p>\n
\u8bf7\u76f4\u63a5\u7ed9\u51fa\u7b54\u6848，\u4e0d\u8981\u89e3\u91ca\u63a8\u7406\u8fc7\u7a0b\u3002
\n"""
\n answer = llm(answer_prompt)<\/p>\n
# \u7b2c\u4e8c\u8f6e：\u81ea\u6211\u68c0\u67e5
\n review_prompt = f"""
\n\u4f60\u73b0\u5728\u5207\u6362\u6210\u4e00\u4e2a\u4e25\u683c\u7684\u5ba1\u7a3f\u4eba\u3002<\/p>\n
\u95ee\u9898：{question}
\n\u4f60\u7684\u56de\u7b54：{answer}
\n\u53c2\u8003\u4e0a\u4e0b\u6587：{context}<\/p>\n
\u8bf7\u68c0\u67e5\u4f60\u7684\u56de\u7b54\u662f\u5426\u5b58\u5728\u4ee5\u4e0b\u95ee\u9898：
\n1. \u5f15\u7528\u4e86\u4e0a\u4e0b\u6587\u4e2d\u4e0d\u5b58\u5728\u7684\u5185\u5bb9？
\n2. \u6570\u5b57\/\u540d\u79f0\/\u65e5\u671f\u7b49\u4e8b\u5b9e\u6027\u5185\u5bb9\u662f\u5426\u6709\u9519\u8bef？
\n3. \u662f\u5426\u5b58\u5728\u660e\u663e\u7684\u903b\u8f91\u77db\u76fe？
\n4. \u662f\u5426\u6709\u65e0\u4f9d\u636e\u7684\u7edd\u5bf9\u5316\u8868\u8ff0？<\/p>\n
\u5982\u679c\u6709\u95ee\u9898，\u8bf7\u7528\u81ea\u7136\u8bed\u8a00\u8bf4\u660e\u95ee\u9898，\u5e76\u7ed9\u51fa\u4e00\u4e2a\u201c\u4fee\u6b63\u5efa\u8bae\u201d；\u5982\u679c\u6ca1\u6709\u95ee\u9898，\u8bf7\u53ea\u56de\u7b54\u201c\u6ca1\u6709\u95ee\u9898\u201d\u3002
\n"""
\n review = llm(review_prompt)<\/p>\n
# \u7b2c\u4e09\u8f6e：\u5982\u679c\u6709\u95ee\u9898，\u6839\u636e\u53cd\u9988\u4fee\u6b63
\n if "\u6ca1\u6709\u95ee\u9898" in review:
\n return answer
\n else:
\n revise_prompt = f"""
\n\u4f60\u521a\u624d\u7684\u56de\u7b54\u5b58\u5728\u95ee\u9898，\u5177\u4f53\u53cd\u9988\u5982\u4e0b：
\n{review}<\/p>\n
\u8bf7\u6839\u636e\u8fd9\u4e9b\u53cd\u9988，\u91cd\u65b0\u7ed9\u51fa\u4e00\u4e2a\u66f4\u51c6\u786e\u3001\u66f4\u8c28\u614e\u7684\u7b54\u6848\u3002
\n\u95ee\u9898：{question}
\n\u4e0a\u4e0b\u6587：{context}
\n"""
\n final_answer = llm(revise_prompt)
\n return final_answer<\/p>\n
\u8fd9\u4e2a\u7248\u672c\u5b9e\u73b0\u6210\u672c\u5f88\u4f4e，\u800c\u4e14\u5bf9\u4f60\u73b0\u6709\u4ee3\u7801\u6539\u52a8\u6700\u5c0f，\u5b8c\u5168\u53ef\u4ee5\u5148\u5728\u4e00\u4e2a\u5173\u952e\u573a\u666f\u8bd5\u70b9：
\n\u6bd4\u5982\u300c\u91cd\u8981\u7684\u5236\u5ea6\u95ee\u7b54\u3001\u5408\u540c\u6761\u6b3e\u89e3\u91ca\u300d\u7b49，\u518d\u6839\u636e\u6548\u679c\u51b3\u5b9a\u662f\u5426\u5347\u7ea7\u5230\u201c\u53cc Agent \u6a21\u5f0f\u300d\u3002<\/p>\n
\u65e2\u7136\u4f60\u524d\u9762\u5df2\u7ecf\u6709：<\/p>\n
\u90a3\u9a8c\u8bc1\u8fd9\u5957\u201c\u81ea\u6211\u53cd\u601d\u201d\u80fd\u529b\u662f\u4e0d\u662f\u503c\u5f97\u4fdd\u7559，\u53ef\u4ee5\u76f4\u63a5\u505a\u4e00\u4e2a\u5c0f\u5b9e\u9a8c：<\/p>\n
\u4f60\u8981\u91cd\u70b9\u5173\u6ce8\u7684\u662f：<\/p>\n
\u5728\u76d1\u63a7\u7cfb\u7edf\u91cc\u53ef\u4ee5\u52a0\u4e24\u7c7b\u6307\u6807：<\/p>\n
\u5982\u679c\u4f60\u5728\u5b9e\u9a8c\u91cc\u770b\u5230：<\/p>\n
\u90a3\u8fd9\u5957\u201c\u81ea\u6211\u53cd\u601d\u80fd\u529b\u201d\u5c31\u975e\u5e38\u503c\u5f97\u5728\u5173\u952e\u4e1a\u52a1\u573a\u666f\u91cc\u63a8\u5e7f\u3002<\/p>\n
\u63a8\u8350\u4e00\u4e2a\u5c3d\u91cf\u201c\u65e0\u75db\u201d\u7684\u63a5\u6cd5：<\/p>\n
\u53ea\u66ff\u6362\u5173\u952e\u8282\u70b9<\/p>\n
\u4fdd\u6301\u5916\u90e8\u63a5\u53e3\u4e0d\u53d8<\/p>\n
\u5148\u7070\u5ea6\u90e8\u5206\u6d41\u91cf<\/p>\n
\u628a\u53cd\u601d\u8fc7\u7a0b\u7684\u5173\u952e\u4fe1\u606f\u5199\u8fdb Trace<\/p>\n
\u8fd9\u56db\u70b9\u5b8c\u6210\u540e，\u4f60\u7684 Agent \u7cfb\u7edf\u5c31\u4ece：<\/p>\n
\u201c\u4f1a\u56de\u7b54\u95ee\u9898，\u4f46\u81ea\u5df1\u4e0d\u68c0\u67e5\u201d<\/p>\n
\u5347\u7ea7\u6210：<\/p>\n
\u201c\u65e2\u4f1a\u56de\u7b54\u95ee\u9898，\u53c8\u4f1a\u5148\u68c0\u67e5\u81ea\u5df1\u8bf4\u7684\u9760\u4e0d\u9760\u8c31\u201d\u3002<\/p>\n
\u8fd9\u4e00\u7bc7\u505a\u7684\u4e8b\u60c5\u662f：\u5728\u4f60\u5df2\u7ecf\u6709\u7684 Agent \u5bb6\u65cf + \u5de5\u4f5c\u6d41 + \u76d1\u63a7 \u57fa\u7840\u4e4b\u4e0a，\u518d\u5f80\u524d\u8fc8\u4e00\u6b65\u2014\u2014\u8ba9 Agent \u62e5\u6709\u4e00\u4e2a\u7b80\u5355\u4f46\u975e\u5e38\u5b9e\u7528\u7684\u80fd\u529b：<\/p>\n
\u5728\u5f20\u53e3\u4e4b\u524d，\u5148\u95ee\u95ee\u81ea\u5df1：
\n\u201c\u6211\u8bf4\u7684\u6709\u4f9d\u636e\u5417？\u6709\u6ca1\u6709\u53ef\u80fd\u5728\u778e\u7f16？\u201d<\/p>\n
\u843d\u5730\u4e0a，\u4f60\u53ef\u4ee5：<\/p>\n
\u524d\u9762\u5df2\u7ecf\u642d\u597d\u4e86：\u80fd\u7528 RAG \u68c0\u7d22\u6587\u6863\u7684\u95ee\u7b54\u94fe\u8def；\u591a Agent \u7684\u201c\u5bb6\u65cf\u5206\u5de5\u201d\u548c\u8c03\u5ea6\u5668；\u8c03\u7528\u76d1\u63a7\u3001Trace\u3001A\/B \u5b9e\u9a8c\u7b49\u5de5\u7a0b\u80fd\u529b\u3002\u4f46\u5728\u771f\u5b9e\u4f7f\u7528\u4e2d，\u4f60\u4e00\u5b9a\u5df2\u7ecf\u9047\u5230\u8fc7\u8fd9\u79cd\u60c5\u51b5：\u660e\u660e\u6587\u6863\u91cc\u6ca1\u6709\u5199，Agent \u5374\u4e00\u672c\u6b63\u7ecf\u5730\u201c\u7f16\u201d\u51fa\u4e00\u6bb5\u5185\u5bb9；\u6570\u5b66\u9898\u3001\u903b\u8f91\u9898\u5076\u5c14\u7ed9\u9519，\u8fd8\u5199\u5f97\u975e\u5e38\u81ea\u4fe1；\u95ee\u201c\u4e3a\u4ec0\u4e48\u201d\u65f6，\u5b83\u4f1a\u51ed\u611f\u89c9\u80e1\u4e71\u63a8\u65ad\u4e00\u4e2a\u7406\u7531\u3002\u6839\u672c\u539f\u56e0\u662f<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[74,4119,50],"topic":[],"class_list":["post-73729","post","type-post","status-publish","format-standard","hentry","category-server","tag-agent","tag-rag","tag-50"],"yoast_head":"\n