Loading ToolBench
Language-isolated Python ReAct runtime for Nino Agent with API-first enterprise agent orchestration, skill management, and MCP server integration
Three tools with strong naming (verb_noun pattern: get_, query_, find_) and comprehensive descriptions (194 - 280 chars, well above 10-char minimum). All parameters have type definitions and descriptions. Input schemas are properly structured with type and description fields. However, output schemas are not documented in the visible source, only inferred from return type names (OrderDetailResult, DataSummaryResult, DataAnomalyResult). Error handling is minimal: GetOrderDetailAsync returns a warnings array for 'not found' but lacks actionable recovery guidance. No pagination support visible despite tools returning potentially large result sets (anomalies capped at 20, but summary groups are unbounded). Tool descriptions are domain-specific and clear but lack explicit 'when to use' guidance for LLM selection.
Finds the lowest negative-margin paid, non-test demo orders for a half-open date range and returns deterministic reason codes.
Gets one demo order with customer resources, supplier resources, payments, refunds, and deterministic totals. Use when an exact order_serial_id is known.
Summarizes paid, non-test demo orders for a half-open date range. Returns deterministic grand totals, including negativeMarginOrderCount, plus groups by main_product_type, channel, or day. Amounts are CNY in the demo dataset; use totals instead of adding groups or counting paginated anomaly rows in the model.
Output schemas not documented. Return types (OrderDetailResult, DataSummaryResult, DataAnomalyResult) are inferred from C# class names but their field structure is not visible in source. LLMs cannot plan downstream tool calls or extract specific fields without documented output schemas.
Error handling lacks recovery guidance. GetOrderDetailAsync returns a warnings array for missing orders, but does not suggest next steps (e.g., 'Try query_summary to find available orders'). Errors should guide the LLM toward resolution.
No pagination or result-limiting guidance for query_summary. Tool description does not state whether groupBy results are capped or paginated. Large result sets risk context window exhaustion.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | <=2025-11-25 | v2 |
Parameter descriptions lack format constraints. startDate and endDate are described as 'YYYY-MM-DD format' in text but no regex pattern or JSON Schema format constraint is visible. LLMs may pass invalid dates.
Tool descriptions lack 'when to use' guidance. Descriptions state WHAT each tool does but not WHEN to call it vs. alternatives. LLMs may struggle to select the right tool when multiple options exist.