Loading ToolBench
Enterprise AI workflow governance platform — ontology layer + MCP + template library. Enables AI-proposed, human-approved workflows with staged writes and approval workflows.
Server has 11 tools with mixed quality. Strengths: all tools have descriptions (10-200 chars, mostly good); tool annotations present (readOnlyHint, destructiveHint, idempotentHint, openWorldHint); clear staging/approval workflow. Weaknesses: 5 tools (query_*, execute_*) are dynamically generated with inferred schemas, cannot verify actual schema completeness in source; parameter descriptions sparse for dynamic tools; no output schema documentation; error handling guidance absent; no pagination/limit enforcement visible for list tools.
批准 staged 动作（批准即应用 transform）
执行 {name} 动作。动态生成的执行工具，基于 ActionType 配置。需要审批的动作在描述中标注（需要审批）。
列出待审批的 staged 动作（含参数、暂存者、before 快照）
查询 {name} 对象。动态生成的查询工具，基于 ObjectType 配置（exposed=true）。
拒绝 staged 动作（必须附拒绝理由）
列出已安装的工作流模板（四段式：对象模型+动作库+工作流+评测集）。看有什么现成能力、或确认 forge 产物是否上架，先查这里。
列出元铸工坊可用的动作类型（按域分组，含中文说明与自主性等级 L1自动/L2人审）。只读，用于发现能做什么。
Dynamic tools (query_*, execute_*) have inferred schemas, actual JSON Schema completeness cannot be verified. Code shows 'params_schema_json or {"type": "object"}' fallback, meaning many execute tools may have empty schemas.
No output schema documentation for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., exec_id from approve_action response). Violates pattern:tool requirement.
No error handling guidance. Tools lack recovery hints (e.g., 'If exec_id not found, call list_pending_approvals first'). Violates pattern:recovery-guide.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 65 | <=2025-11-25 | v2 |
执行元铸工坊的一个动作（写操作）。L1 自动生效；L2 只会进入 staged 待审，等人批准。动作名与参数 schema 用 yuanzhu_actions 查。
一句话铸造工作流模板（AI 生成四段式→结构校验→冒烟评测→全过自动上架）。描述要含：谁/做什么/产出什么。耗时约 30-60 秒。
列出元铸工坊待审的 staged 动作（AI 提交的写操作待人拍板）。只读。
按对象类型查询元铸工坊本体层的对象（只读）。对象类型可先用 yuanzhu_actions 旁的 /api/objects/types 发现，工具名为 query_{domain}_{type} 小写。
Dynamic query tools (query_*) lack pagination/limit enforcement. No visible max result cap or next_cursor pattern. Large result sets risk context window exhaustion.
Parameter descriptions for dynamic tools are generic ('Schema derived from ActionType.params_schema_json'). LLMs cannot infer valid values or constraints without concrete examples or enum declarations.