Loading ToolBench
A desktop pet application with AI agent capabilities, supporting Live2D animation, chat, and web tools plugin system
Two tools with complete JSON Schema definitions and Chinese descriptions. Both tools have proper input schemas with required fields and type declarations. However, descriptions lack depth on WHEN to use each tool vs alternatives, error recovery guidance is minimal, and output schemas are not documented. Parameter descriptions are present but generic. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite fetch_url and search_web being read-only operations. Error handling returns basic messages without recovery guidance.
获取指定 URL 的网页内容。会自动提取页面正文文本，去除 HTML 标签、脚本和样式。适合获取文章、文档、API 响应等内容。
使用搜索引擎搜索关键词，返回搜索结果列表（标题、链接、摘要）。当需要查找最新信息、回答事实性问题时使用。
Output schemas not documented. LLMs cannot infer what fields fetch_url and search_web return, forcing them to guess at response structure and plan downstream operations blindly.
Tool annotations missing. Both tools are read-only (risk: READ_ONLY) but lack readOnlyHint annotation in schema. This prevents agents from optimizing retry logic and understanding safety properties.
Error messages lack recovery guidance. 'HTTP 404' and 'Unknown tool' tell the agent nothing about what to do next. Should include suggestions like 'Try a different URL' or 'Check the tool name'.
Parameter 'count' in search_web lacks type constraint in description. Description says 'default 5, most 10' but does not state minimum=1, maximum=10 explicitly. LLMs may pass invalid values.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | <=2025-11-25 | v2 |
Descriptions do not explain WHEN to use each tool vs alternatives. 'fetch_url' vs 'search_web' distinction is unclear, when should an agent choose one over the other? No guidance on prerequisites or dependencies.