当前位置：首页 > news >正文

16.修正 LangGraph Agent 的路由层，让 Router 真正只负责选工具

news 2026/4/16 1:23:35

1. 今天做了什么

昨天的测试日志有暴露出问题：路由在选择工具和提取用户问题的时候回在半路就将用户的问题给回答了，所以今天要做的是给路由增加约束条件避免中途回答用户问题。

2. 核心变化

首先是在node.py中增加两个小工具，这两个小工具一个负责给出更规范的json字典，另一个负责给工具选择结果加强约束，如果是rag、llm、time这三类，默认都把原始 query 继续传下去。

defclean_json_text(text:str)->str:text=text.strip()iftext.startswith("```json"):text=text.removeprefix("```json").strip()eliftext.startswith("```"):text=text.removeprefix("```").strip()iftext.endswith("```"):text=text.removesuffix("```").strip()returntextdefnormalize_decision(decision:dict,query:str,valid_tool_names:set[str])->dict:ifnotisinstance(decision,dict):return{"tool":"llm","input":query}tool=str(decision.get("tool","")).strip().lower()tool_input=str(decision.get("input","")).strip()iftoolnotinvalid_tool_names:return{"tool":"llm","input":query}# 关键规则：# 对 rag / llm / time 这三类，默认都把原始 query 继续传下去，# 不允许 router 自己现编一个答案塞进 input。iftoolin{"rag","llm","time"}:return{"tool":tool,"input":query}# calculator 允许保留模型抽出来的表达式iftool=="calculator":ifnottool_input:return{"tool":"calculator","input":query}return{"tool":"calculator","input":tool_input}return{"tool":"llm","input":query}

3. 一个关键问题

问题是什么：路由会提前回答问题原因是由于工具选择node的约束不够强，所以导致大模型自己回答了问题。
怎么解决的：做更强的提示词工程，增加一些约束条件，约束模型的返回结果。

defbuild_choose_tool_node(tools:list[dict[str,Any]]):valid_tool_names={t["name"]fortintools}defchoose_tool_node(state:AgentState)->AgentState:query=state["query"]logger.info(f"[choose_tool_node] query:{query}")tool_desc="\n".join([f"{t['name']}:{t['description']}"fortintools])prompt=f""" You are a tool router. Your job is ONLY to choose the best tool and prepare its input. Do NOT answer the user's question. Do NOT rewrite the user's question into an answer. Return JSON only. Available tools:{tool_desc}Rules: 1. You must return exactly one JSON object. 2. JSON format: {{"tool": "...", "input": "..."}} 3. tool must be one of: rag, calculator, time, llm 4. For rag, llm, and time: - input should stay the same as the user's original question - do not invent a new sentence 5. For calculator: - input should be the math expression only if you can extract it 6. Do not include markdown, explanations, or code fences. User question:{query}"""content=""try:response=client.chat.completions.create(model=CHAT_MODEL,messages=[{"role":"user","content":prompt}])content=response.choices[0].message.content cleaned=clean_json_text(content)raw_decision=json.loads(cleaned)decision=normalize_decision(raw_decision,query,valid_tool_names)logger.info(f"[choose_tool_node] raw decision:{raw_decision}")logger.info(f"[choose_tool_node] normalized decision:{decision}")return{"decision":decision}exceptExceptionase:logger.exception("choose_tool_node failed")return{"decision":{"tool":"llm","input":query},"error":f"choose_tool_node failed:{str(e)}"}returnchoose_tool_node