Loading ToolBench
Open the browser and visit the specified website address
Single tool 'open-browser' has a basic description and minimal schema. The tool name lacks a clear action verb (should be 'open_browser' or 'launch_browser'). Input schema is present but parameter description is generic and includes example values (anti-pattern). No output schema documented. No error handling guidance, the catch block returns a generic failure message without actionable recovery steps. No parameter validation, no idempotency guarantees, and no security considerations for arbitrary URL input.
Open the browser and visit the specified website address
Tool name uses hyphen instead of underscore and lacks clear action verb. 'open-browser' should be 'open_browser' or 'launch_browser' to match verb_noun convention and improve LLM parsing.
Parameter description includes example values ('https://www.baidu.com, https://www.google.com') which LLMs tend to reuse literally. Replace with formal constraints (URL format, protocol requirements) instead.
No output schema documented. LLMs cannot plan downstream actions or extract structured data. Document what fields the response contains and their types.
Error handling returns generic 'The browser failed to open' without actionable recovery guidance. Should indicate whether the error is retryable, what caused it (invalid URL, browser not installed, permission denied), and what the LLM should do next.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 38 | 2026-07-28+ | v2 |
No input validation. Tool accepts any string as URL without checking format, protocol, or safety. Arbitrary URLs could be passed; tool should validate URL format and optionally restrict to http/https protocols.