Loading ToolBench
Single tool with basic schema and descriptions. Tool name 'capture_screen' is verb-noun compliant. Input schema is present with Zod validation and type definitions (string, boolean). However, descriptions are minimal (10-20 chars for tool, 30-60 chars for params), falling below the 50-200 char LLM-optimized baseline. No output schema documented. Error handling returns JSON but lacks recovery guidance or error classification. No parameter constraints (enums, ranges). Tool is directly visible in code (not inferred), but overall quality is below production baseline.
全屏截图：捕获整个屏幕的画面
Tool and parameter descriptions are too brief (10-60 chars). LLM-optimized descriptions should be 50-200 chars, explaining WHAT the tool does, WHEN to use it, and any prerequisites. Current descriptions lack context for LLM selection.
Output schema is not documented. Tool returns JSON with success/error fields, but LLMs cannot infer the response structure. Document the return type (success: boolean, filePath: string, fileName: string, captureTime: string, options: object).
Error responses lack recovery guidance. When screencapture fails, the tool returns a generic error message. Should include actionable next steps: 'Check if ~/Desktop/Screenshots/ is writable' or 'Verify screencapture command is available on this macOS version'.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 48 | 2026-07-28+ | v2 |
No parameter constraints or ranges. 'outputPath' accepts any string; no validation that it's a valid directory path or that it won't cause path traversal. 'includeCursor' and 'showPreview' are booleans with defaults, but no explanation of side effects.
Tool description in Chinese ('全屏截图：捕获整个屏幕的画面') may not be parsed correctly by English-language LLMs. Descriptions should be in English for maximum compatibility.