Loading ToolBench
Production-grade proxy layer for Kingdee K3Cloud WebAPI, providing secure MCP interface with credential isolation, read-only/read-write modes, write operation confirmation gates, full audit logging, query protection, and team token authentication.
Server has 8 tools with visible schemas and descriptions in Chinese. Naming follows verb_noun pattern (query_, view_, preview_, confirm_). Schemas are present with type definitions and descriptions. However, descriptions are brief (20-80 chars), lack WHEN/WHY context, and parameter descriptions are minimal. No output schemas documented. Error handling is absent, no guidance on retryability, user-fixable vs fatal errors, or recovery paths. Write tools lack explicit confirmation flow documentation in descriptions. Security model (credential isolation, audit logging) is well-implemented in code but not surfaced in tool definitions.
【写】确认执行之前排队的操作（preview_save_bill、preview_submit_bills、preview_audit_bills、preview_unaudit_bills、preview_delete_bills）。
【写】审核单据：先预览排队，需 confirm_action 确认。
【写】删除单据：先预览排队，需 confirm_action 确认。
【写】新建/修改单据：只生成预览并排队，需 confirm_action 确认后才真正保存。
【写】提交单据：先预览排队，需 confirm_action 确认。
【写】反审核单据：先预览排队，需 confirm_action 确认。
通用单据查询（只读）。form_id 如 BD_MATERIAL(物料)、SAL_SaleOrder(销售订单)、PUR_PurchaseOrder(采购订单)、STK_Inventory(即时库存)。filter_string 用金蝶类 SQL语法，如 FDocumentStatus = 'C' 表示已审核。
No output schemas documented. LLMs cannot predict response structure or plan downstream tool calls. query_bill, view_bill, and all preview_* tools lack documented return types.
Descriptions lack WHEN/WHY context and are under 100 chars. view_bill ('查看单据完整详情（只读）') does not explain when to call it vs query_bill, or what fields are returned. Descriptions should be 50-200 chars with actionable guidance.
No error handling guidance. Tools do not document retryability, user-fixable errors, or recovery paths. E.g., confirm_action raises ValueError for missing ticket but does not guide LLM on what to do next.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | <=2025-11-25 | v2 |
查看单据完整详情（只读）。
Confirmation flow not explicit in tool descriptions. preview_* tools return a ticket but descriptions do not clearly state 'Call confirm_action with this ticket to execute' or explain the two-phase commit pattern.
query_bill limit parameter defaults to 100 but description does not explain the 1000-row hard cap or why it exists. Unbounded queries could timeout or exhaust context.