Loading ToolBench
Two tools with basic schemas and descriptions, but significant gaps in parameter documentation and output schema clarity. get_greeting has a single well-described parameter; calculate has three parameters but lacks descriptions for 'operation' enum values and output structure. No error handling guidance, no idempotency hints, and descriptions are minimal (under 100 chars). Tool names follow verb_noun convention but lack context about when to use each. Output schemas are not documented, LLMs cannot plan downstream calls or extract structured data reliably.
Perform a calculation based on the given operation and operands
Get a greeting for the given name
Output schemas not documented. LLMs cannot infer what fields to expect from tool responses, blocking downstream tool chaining and data extraction.
calculate tool lacks parameter descriptions for 'operation' enum. LLMs cannot determine valid values (add, subtract, multiply, divide) without explicit enum constraint or description.
No error handling guidance. Tool descriptions do not explain what happens on invalid input (e.g., divide by zero, invalid operation) or how to recover.
Tool descriptions are too brief (under 100 chars). They lack context about WHEN to use each tool, prerequisites, or what the tool returns.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | <=2025-11-25 | v2 |
No idempotency or state-modification hints. Descriptions do not clarify whether tools are safe to retry or have side effects.