Loading ToolBench
An educational web application demonstrating AI/LLM capabilities across 8 learning levels, from pure LLM chat to MCP (Model Context Protocol) integration. Built with ASP.NET Core Blazor, it showcases Function Calling, Agents, RAG, and tool integration patterns.
All 4 tools have basic descriptions in Chinese, but lack depth and LLM-optimization. Input schemas are present with type definitions, but parameter descriptions are minimal (10-20 chars). No output schemas documented. Tool names lack English verb-noun clarity (Chinese descriptions only). No error handling guidance, no pagination support, no idempotency hints. GetCurrentTime has zero parameters but no output schema. SubmitReturnRequest is a write operation with no confirmation/dry-run pattern. SearchDocuments simulates RAG but returns unstructured mock data without pagination or result limits.
取得目前時間
查詢訂單狀態
搜尋內部文件庫 (模擬 RAG)
提交退貨申請
No output schemas documented. LLMs cannot predict response structure or plan downstream calls. GetOrderStatus returns JSON string; SearchDocuments returns mock dict; GetCurrentTime has no documented return type.
Parameter descriptions are too brief (10-20 chars). 'orderId' → '訂單編號' lacks format guidance (e.g., 'Format: ORD-XXXXXX'). 'reason' → '退貨原因' does not explain length limits or required detail level.
SubmitReturnRequest is a destructive write operation with no confirmation, dry-run, or idempotency hint. Agents may retry and create duplicate return requests.
SearchDocuments returns hardcoded mock data without pagination, result limits, or relevance scoring. No guidance on when to call it or how many results to expect.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
Tool names are in English but descriptions are in Chinese only. LLMs trained on English will struggle to infer intent from names alone without bilingual descriptions.