Three tools with complete schemas and descriptions, but significant gaps in parameter documentation and error handling. Tool names use action verbs (app_management, system_management, lww_management) but are overly generic and combine multiple responsibilities. Descriptions are present (avg ~80 chars) but lack actionable context for LLM selection. Parameters have types and enums but lack descriptions for nullable fields. No output schema documentation. Error handling and recovery guidance absent. Follows basic structure but falls short of production-grade quality.

• Split app_management into five separate tools: list_apps(), get_app_status(app), start_app(app, script), stop_app(app, script), delete_app(app). Each tool has one clear responsibility and a verb-noun name LLMs can parse instantly.
• Add descriptions to all parameters, especially nullable ones. E.g., 'app: App folder name (e.g., "my-ai-app"). Required for status, start, stop, delete; omit for list.' Include format hints and constraints.
• Document output schemas for each tool. E.g., app_management returns {status: 'running'|'stopped', pid?: number, error?: string}. lww_management(action='devices') returns {devices: [{host, port, online}]}. This enables LLM to extract and chain results.
• Add error handling guidance. E.g., 'If Pinokio is offline, start_app returns {error: "Pinokio not running"}. Call system_management(action="ping") first to verify connectivity.' Categorize errors as retryable vs user-fixable.
• Rename tools for clarity: ping_pinokio, get_pinokio_info, search_localhost_services, discover_pinokio_devices, start_remote_app. Avoid generic names and domain jargon.