Server has 7 tools with schemas and descriptions, but quality is inconsistent. Descriptions are present but often terse (10-50 chars) and lack actionable context for LLM selection. Schemas are visible in server.py but parameter descriptions are minimal or missing context about constraints, formats, and dependencies. No error handling guidance, no output schema documentation, and no tool annotations (readOnlyHint/destructiveHint). Chinese descriptions reduce clarity for English-speaking LLM agents. Tools like 'act' combine multiple concerns (click/input/press/scroll/enter) into one parameter-heavy tool, violating single-responsibility principle.

• Translate all descriptions to English and expand to 50-150 characters. Include WHEN to use the tool and WHAT it returns. E.g., 'list_targets: Enumerate all open windows and detect available interaction channels (UIA, CDP, OCR). Call this first to identify the target window handle.'
• Split 'act' into separate tools: click_element(target_id), input_text(target_id, text), press_keys(keys), scroll_region(x, y, width, height, delta), enter_key(). Each tool has one clear responsibility and LLM can compose them naturally.
• Document output schemas for all tools. E.g., snapshot returns {source: string, elements: [{id: string, type: string, text: string, ...}]}. This enables LLMs to extract target_id for downstream act() calls.
• Add enum constraints to parameters. Declare prefer as enum: ['auto', 'cdp', 'uia', 'ocr']. Add bounds to max_items (1-500), timeout (0.1-60), x/y/width/height (0-9999). Include units in descriptions: 'timeout in seconds'.
• Add error recovery guidance. E.g., act() returning {ok: false, reason: 'id_missing'} should suggest 'Call snapshot() again to refresh element IDs' or 'Element may have been removed; try wait_change() to detect UI updates'.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry