Loading ToolBench
Two tools with minimal definitions. Both have descriptions (10-20 chars, below baseline 194 chars), and schemas are present but sparse. 'sum' has typed parameters with descriptions; 'members' has empty input schema. No error handling, no output schema documentation, no parameter constraints or validation guidance. Naming is generic ('sum', 'members'), lacks action verbs and clarity. No composition hints or chaining IDs. Descriptions are too brief to guide LLM selection.
Get members
Sum two numbers
Tool descriptions are too brief (10-20 chars vs. baseline 194 chars). 'Sum two numbers' and 'Get members' lack context for LLM selection, prerequisites, or when to use each tool.
Tool names lack action verbs. 'sum' and 'members' are generic nouns, not verb_noun patterns like 'calculate_sum' or 'list_members'. LLMs cannot infer intent from names alone.
'members' tool has empty input schema ({}). No parameters documented, no constraints, no guidance on what data it returns or when to call it.
No output schema documented for either tool. LLMs cannot plan downstream calls or extract structured data. Responses are text blobs requiring parsing.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | 2026-07-28+ | v2 |
No error handling or recovery guidance. If 'sum' receives non-numeric input or 'members' fails to load JSON, no actionable error message guides the LLM to retry or correct input.