Loading ToolBench
ZenoAgent provides 7 tools with mostly complete schemas and descriptions in Chinese. Tool naming follows verb_noun patterns (system_ask_user_question, system_fetch_url, system_load_skill, system_rag_query, system_resolve_tools, system_sub_agent, system_todo_write). All tools have non-empty descriptions (ranging 80-400+ chars, within the 10-1024 baseline). Input schemas are present for all tools with proper JSON Schema structure including type, properties, required fields, and descriptions. Error handling guidance is minimal; tools do not provide recovery instructions or error categorization. Parameter descriptions are present but some lack constraint detail (e.g., system_ask_user_question's 'options' array lacks min/max items guidance). Tool descriptions are oriented toward the LLM's decision-making ('when to call this tool') but lack actionable guidance on error recovery.
向用户发起提问以获取必要信息。支持四种形式：SINGLE_SELECT（单选）、MULTI_SELECT（多选）、INPUT（自由输入）、PREVIEW（预览确认）。系统会自动为选项补充「其他」选项。当任务需要用户提供选择或补充信息时调用。
获取指定 URL 网页的正文内容。注意：不支持需要 JavaScript 动态渲染的页面。参数 truncate 控制输出长度：truncate=true（默认）：返回前 5000 字。适用于【快速了解页面概要、判断内容是否相关、提取关键摘要信息】等场景，大多数情况下应优先使用。truncate=false：返回完整正文（可能很长）。仅在以下情况使用：1. 用户明确要求「完整内容」或「全文」；2. 截断版本已读取但信息不完整，需要继续获取剩余内容；3. 需要精确引用原文的特定段落或数据。
按需加载指定 Skill 的完整内容。当你需要使用某条技能（Skill）时，先调用本工具传入技能 ID，系统将返回该技能的完整操作指南，你再按照指南执行。
在知识库中语义检索相关文档。当需要查询业务知识、文档内容、历史记录时调用此工具。
按需加载指定 MCP 工具的完整定义（参数 schema）。当你判断需要使用某些工具但当前无法直接调用时（工具列表中提示"按需加载"），先调用本工具传入工具名列表，下一轮推理中这些工具将出现在可调用列表中。
启动子 Agent 处理复杂多步子任务（功能开发中）。
NO DOCUMENTED OUTPUT SCHEMAS. Tools define input schemas but nowhere in the provided code or documentation are return types, field structures, or response examples documented. LLMs need to know what fields to expect so they can plan downstream tool calls and extract the right data.' This is a critical omission that prevents downstream tool chaining.
MINIMAL ERROR HANDLING GUIDANCE. Tool descriptions state what the tool does but do NOT explain error recovery paths, retryability, or categorization. Try search_users() with a partial name." A raw error code or stack trace gives the agent nothing to act on.' No tool description addresses failure modes or recovery strategies.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 63 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 30 | - | v1 |
任务记事本工具。当用户指令较为复杂、且可拆分为多个相对独立的步骤时，使用此工具管理任务清单。仅在任务确实复杂（3步以上）且各步骤相对独立时才创建 Todo，简单任务无需创建。每完成一个步骤后，立即调用 update 将对应 Todo 标记为 completed。发现某步骤不再需要执行时，将其标记为 cancelled。支持三种操作（action）：create - 批量创建 Todo 条目，通过 todos 参数传入任务列表；update - 更新指定 Todo 的状态，通过 updates 参数传入变更列表；list - 查询当前所有 Todo 条目（正常情况下无需主动调用，系统提示词中已包含）
VAGUE PARAMETER CONSTRAINTS. Several parameters lack explicit min/max or enum bounds. Example: system_ask_user_question 'options' array has no minItems or maxItems constraint; system_rag_query 'knowledgeIds' lacks bounds; system_todo_write 'priority' integer has no documented enum values (stated as 1/2/3 in description but not in schema).
INCOMPLETE TOOL COMPOSITION GUIDANCE. system_resolve_tools and system_load_skill are discovery/utility tools but descriptions do not explain typical patterns for chaining them with other tools or when to call them in relation to other tools.
SECURITY: system_sub_agent marked WRITE but no permission gating documented. Verify the calling user/agent has authority before executing.' Launching sub-agents could escalate privileges, no mention of permission validation in description or visible in code.