Loading ToolBench
Single tool with basic structure but significant quality gaps. Tool name 'execute_gas_code' lacks action clarity (execute is vague; 'run_gas_function' would be clearer). Description is adequate (89 chars, within 10-1024 range) but lacks WHEN to use it and prerequisites. Input schema present with 4 parameters, all with descriptions, but schema lacks type constraints (no enums, no min/max for arrays). Output schema not documented, LLM cannot predict response structure. Error handling returns generic error dicts without recovery guidance. No parameter validation rules stated (e.g., code length limits, function_name format). Missing security guidance on code injection risks.
Execute Google Apps Script (GAS) code in a secure virtual environment.
Output schema not documented. LLM cannot predict response structure (success vs error format, field names, types). Breaks downstream tool chaining.
Tool name 'execute_gas_code' is vague. 'execute' does not clearly signal the action. Rename to 'run_gas_function' or 'invoke_gas_script' for clarity.
No input validation constraints. Parameters lack enums, min/max, length limits, or regex patterns. 'code' parameter unbounded, LLM could pass gigabytes of code.
Error responses are generic dicts without recovery guidance. 'HTTP request error: ...' or 'Execution error: ...' tells LLM nothing about retry eligibility or next steps.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 40 | 2026-07-28+ | v2 |
GAS_API_KEY exposed as environment variable without mention of secret injection pattern. No audit trail or permission gates documented for code execution (high-risk operation).