24 tools with basic schemas and descriptions, but significant quality gaps. Most tool descriptions are 10-30 chars (well below 50-200 char baseline for A+ tools). Parameter descriptions exist but are minimal (e.g., 'X coordinate' for x param). No output schemas documented. No error handling guidance. No tool annotations (readOnlyHint/destructiveHint). Many tools lack actionable context for LLM selection. Naming is verb-forward (good), but descriptions do not explain WHEN to use each tool or what it returns.

• Expand tool descriptions to 50 - 150 characters. Include WHAT the tool does, WHEN to use it (vs similar tools), and what it returns. E.g., 'Click at screen coordinates. Use after screenshot() to interact with UI elements. Returns success or error if coordinates are off-screen.'
• Document output schemas for all tools. E.g., screenshot returns {success: bool, imageBase64: string, width: number, height: number}. listDisplays returns {displays: [{id: number, width: number, height: number}]}. This enables LLM planning.
• Add tool annotations: readOnlyHint=true for screenshot, listDisplays, getCursorPosition, etc. destructiveHint=true for click, type, drag, openApp, etc. This guides LLM retry logic and safety reasoning.
• Add parameter constraints and ranges. E.g., x and y should document screen bounds (0 to display width/height). jpegQuality should be 1 - 100. keySequence should list valid key names. Use enum for button parameter (left|right|middle).
• Add error handling guidance. E.g., 'If coordinates are outside display bounds, returns error with available bounds. Retry with adjusted coordinates.' This enables LLM self-correction without extra discovery calls.

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

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