Loading ToolBench
A Model Context Protocol server for executing browser automation tasks
Single tool with minimal definition quality. Tool name 'execute_browser_task' lacks action clarity (execute is vague; better: 'automate_browser_task' or 'run_browser_automation'). Description is 28 characters, below the 34-char p10 baseline and provides no context on WHEN to use, WHAT it returns, or prerequisites. Input schema present but parameter 'task' lacks type validation, format constraints, or length bounds. No output schema documented, LLM cannot predict response structure. No error handling guidance. Tool performs a WRITE operation (browser automation) but lacks idempotent hints, confirmation patterns, or recovery guidance. Chinese descriptions reduce accessibility for international teams.
执行浏览器自动化任务
Tool name 'execute_browser_task' is vague. 'Execute' does not clearly signal browser automation intent. Rename to 'automate_browser_task' or 'run_browser_automation' to match verb_noun pattern and improve LLM discoverability.
Description '执行浏览器自动化任务' (28 chars) is below baseline (p10=34 chars) and lacks critical context: WHEN to use this tool, WHAT it returns, prerequisites (e.g., browser setup), and whether it modifies state. Expand to 50 - 150 chars with actionable guidance.
Parameter 'task' has no type validation, format constraints, or length bounds. LLM may pass arbitrarily long or malformed strings. Add minLength, maxLength, and pattern constraints. Document expected format in description (e.g., 'Natural language task description, 10 - 500 chars').
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 36 | 2026-07-28+ | v2 |
No output schema documented. LLM cannot predict response structure (is it plain text? JSON? HTML?). Document return type: e.g., 'Returns object with fields: success (boolean), output (string), error (string, if failed).'
Tool performs WRITE operations (browser automation) but lacks error recovery guidance, idempotent hints, or confirmation patterns. Error handling returns raw exception messages ('执行出错: ...') instead of actionable guidance. Add error classification (retryable vs. fatal) and recovery steps.