Why do some of my large model tool call requests not enter the tool but return directly?为什么有时我的大模型的工具调用请求没有进入工具，而是直接返回

In most cases, when I explicitly tell the model which tool to use, the model can accurately complete the call, but there are a few cases where the tool_resp request returned by the model is not processed by springAi and returned directly as the model, such as the following:
<tool_call>
{"name": "生成数据", "arguments": {"planId": "0920163013", "userName": "234234", "creditDocInfo": "{"creditType":"FSI","creditDesc":"FDSGV分","queryKey":"PHONE,ID_NO","subSysNames":"CRE-SERVICE,CRE-BATCH","reqType...

What is the problem? Why is the tool call in the <tool_call> tag not actually understood and executed by SpringAi?

在大部分情况下，当我明确告诉了模型应该使用哪个工具，模型都能精准的完成调用，但是存在少部分情况模型返回的tool_resp请求没有被springAi处理，直接作为模型的返回了，例如下面这样：
<tool_call>
{"name": "生成数据", "arguments": {"planId": "0920163013", "userName": "234234", "creditDocInfo": "{\"creditType\":\"FSI\",\"creditDesc\":\"FDSGV分\",\"queryKey\":\"PHONE,ID_NO\",\"subSysNames\":\"CRE-SERVICE,CRE-BATCH\",\"reqTypes\":\"SINGLE_ASYNC,BATCH\",\"interfaceId\":\"447172\",\"creditResult\":[{\"fieldName\":\"scoreid\",\"fieldFullName\":\"fsi.scoreid\",\"fieldDesc\":\"系统产生的唯一评分编号\",\"fieldType\":\"C(120)\"},{\"fieldName\":\"score\",\"fieldFullName\":\"fsi.score\",\"fieldDesc\":\"评分结果\",\"fieldType\":\"N(3)\"},{\"fieldName\":\"reason\",\"fieldFullName\":\"fsi.reason\",\"fieldDesc\":\"原因码\",\"fieldType\":\"C(20)\"}],\"releasePlanId\":\"105857\",\"releasePlanName\":\"【2025.11.18-2099.11.20】发布计划1118\",\"storyId\":\"196862\",\"version\":\"2025.11.18\",\"branchNo\":\"25.11.18\",\"sysStoryInfos\":[{\"storyId\":202210,\"storyName\":\"test数据源接入需求-CDPS-CSERVICE\",\"subSystemId\":\"6307\",\"subSystemName\":\"CDPS-CSERVICE\"},{\"storyId\":202222,\"storyName\":\"test数据源接入需求-CRE-SERVICE\",\"subSystemId\":\"6048\",\"subSystemName\":\"CRE-SERVICE\"},{\"storyId\":202223,\"storyName\":\"test数据源接入需求-DVS-RISKADSDWH\",\"subSystemId\":\"6166\",\"subSystemName\":\"CNCBDP-RISKADSDWH\"}],\"attachInfos\":[{\"fileName\":\"GDGGB源需求文档.docx\",\"fileUrl\":\"http://uat.233.234.com/proxy/get_file/1c343sdf48869a7b09e4a5259426.docx\"},{\"fileName\":\"接口文档.xlsx\",\"fileUrl\":\"http://uat.dfg.sdffd.com/proxy/get_file/076sdf486c3b12037df8593.xlsx\"}]}"}}}
</tool_call>

请问这是什么问题？为什么这个<tool_call>标签的工具调用没有真正被SpringAi理解和执行？

Start with Spring AI's handling of model tool-call responses, using the shown <tool_call> payload as the reproduction input. Confirm whether this response format is supported and record the required model or response format, or a reproducible failure, as the completion criteria.