Loading ToolBench
JavaScript SDK for interacting with the Toolbox service via the Model Context Protocol (MCP). Provides both a core SDK (@toolbox-sdk/core) and an ADK-compatible wrapper (@toolbox-sdk/adk) for loading and executing tools from a remote Toolbox server.
Scoring was not performed
Tool definition is a framework wrapper class found only in test fixtures (packages/toolbox-adk/test/test.tool.ts), not a concrete, production-ready tool with domain-specific logic. The 'ToolboxTool' is an adapter that delegates to an unknown wrapped CoreTool.
Tool description ('An adapter class that wraps a CoreTool from @toolbox-sdk/core to make it compatible with the @google/adk BaseTool interface. Runs tools by delegating calls to the wrapped coreTool.') is generic, framework-focused, and provides no information about WHAT the tool actually does, WHEN to call it, or WHAT it returns. No action verb. Non-actionable for LLM tool selection.
Input schema is minimally defined as a generic object with a 'request' parameter of type 'object' containing 'RunAsyncToolRequest from the ADK agent containing args and toolContext'. No specific parameter names, types, enums, or constraints documented. No parameter descriptions present.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 43 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 16 | - | v1 |
No output schema documented. The runAsync() method returns 'Promise<unknown>' with no documentation of expected return structure, fields, or format.
No error handling or recovery guidance. The runAsync() method has no try-catch, no error categorization, and no guidance for what the LLM should do if execution fails.
Tool name 'ToolboxTool' does not follow verb_noun naming convention. Generic framework name with no action verb (get_, create_, search_, etc.). Does not convey what the tool does.
The wrapped CoreTool's actual schema (zodSchema) is converted via ConvertZodToFunctionDeclaration() inside _getDeclaration(), but this conversion is not visible in the source. The input schema provided in the evaluation is generic and undocumented.