A Spring Boot-based agent framework with ReAct pattern, memory management, RAG integration, and MCP client support
Both tools have descriptions in Chinese, but lack critical schema documentation. Search tool has only a single 'query' parameter with minimal type info. RAGTool similarly has only 'query' parameter. Neither tool shows explicit input schema definitions in the source code provided. Output schemas are completely undocumented. Tool names lack action verbs (Search vs search_web, RAGSearch vs rag_retrieve). No parameter descriptions visible beyond the tool-level description. Error handling is generic ('知识库中未找到相关信息'). The McpTool wrapper shows tool invocation but does not reveal the actual schema registration or validation logic.
此工具用于在知识库中搜索特定知识或文档。当问题涉及可能存储在RAG系统中的特定领域知识(例如:健身基础、公司政策)时,此工具非常有用。
一个基于SerpApi的实战网页搜索引擎工具。它会智能地解析搜索结果,优先返回直接答案或知识图谱信息。当用户询问实时信息、新闻或具体事实时使用此工具。
No input schemas visible in source code. Both tools accept only 'query' parameter but no JSON Schema definitions, type constraints, or parameter descriptions are shown.
Tool names lack action verbs. 'Search' and 'RAGSearch' do not follow verb_noun convention (e.g., search_web, rag_retrieve). LLMs rely on verb-first names to infer intent.
No output schema documentation. Tools return strings but expected fields, structure, and format are not documented. LLMs cannot plan downstream calls or extract structured data.
Parameter 'query' lacks description. The rubric requires every parameter to have a non-empty description explaining what it controls, expected format, and constraints. Only tool-level descriptions exist.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 28 | <=2025-11-25 | v2 |
Error handling is generic and non-actionable. RAGTool returns '知识库中未找到相关信息' (knowledge base not found) with no recovery guidance. McpTool catches exceptions but returns raw error messages without classification (retryable vs fatal) or next-step guidance.