Loading ToolBench
Webサイトの自動操作とデータ抽出を実行するサーバーです。URLと一連のアクションを指定してください。
Single tool 'execute_web_runner' with complete JSON Schema input definition, but significant deficiencies in naming clarity, output documentation, and error handling. Tool name is compound (execute + web_runner) and lacks the verb_noun pattern. Description is in Japanese and minimal (75 chars when translated). Schema is well-structured with proper types and nested objects, but output schema is completely undocumented, the tool returns 'str' (JSON string) with no specification of response structure. No tool annotations present. Error handling exists but lacks recovery guidance and actionable messaging. Input validation constraints are minimal.
指定されたURLとアクションリストに基づいてWebブラウザ自動化タスクを実行し、結果をJSON文字列として返します。
Tool name lacks clarity and action verb. 'execute_web_runner' is compound and vague. Should follow verb_noun pattern like 'automate_web_task' or 'run_browser_automation' to clearly signal intent.
Output schema completely undocumented. Tool returns str (JSON string) but there is no documentation of the JSON structure, fields, or format. LLMs cannot parse or reason about returned data without an explicit schema.
Description is minimal and in Japanese (75 chars when translated: 'Execute web browser automation tasks based on specified URL and action list, returning results as JSON string'). Does not explain WHEN to use this tool, what prerequisites exist, or what types of actions are supported.
Error handling lacks recovery guidance. ToolError messages reference server logs but do not provide actionable next steps. No distinction between retryable vs. fatal errors. Missing classification pattern.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 43 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 41 | - | v1 |
No tool annotations present. Tool performs destructive/write operations (browser automation, form submission, clicks) but lacks destructiveHint or idempotentHint annotations to guide agent planning.
ActionStep items accept free-form 'action' string (e.g. 'click', 'input', 'get_text_content') with no enum constraint. LLM can hallucinate invalid action names. Should constrain to known set: ['click', 'input', 'get_text_content', 'switch_to_iframe', 'select_option', 'get_attribute', 'get_all_attributes', 'wait', ...].
No maximum result size or pagination documented. If browser automation returns large HTML content or long lists, response could exceed context limits. No guidance on handling large outputs.