Loading ToolBench
Single tool 'add' has critical definition gaps. Input schema present with Zod types (a, b as numbers) but parameters lack descriptions entirely (null values). Tool itself has no description visible in code. Output is unstructured text concatenation rather than typed fields. No error handling, validation guidance, or recovery paths. Naming is generic (verb-less); 'add' alone doesn't convey domain context. Follows minimal MCP patterns.
Tool 'add' has no description. LLMs cannot determine when to select this tool or what it does beyond the name.
All input parameters (a, b) lack descriptions (null values in schema). LLMs cannot infer parameter meaning or constraints.
Output is unstructured text (String(a + b)). No typed response schema documented. LLMs cannot parse or chain results.
Tool name 'add' is generic and verb-less. Should be 'calculate_sum' or 'add_numbers' to clarify intent and domain.
No error handling or validation. Invalid inputs (non-numbers, NaN, Infinity) will produce silent failures or garbage output.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 34 | 2026-07-28+ | v2 |