Critical gaps across all dimensions. The 'execute' tool has a basic description but lacks output schema documentation. Dynamic tools (command, list, get, get_application) are inferred from generator.ts rather than explicitly registered with full schemas visible in source. Parameter descriptions exist but are generic and lack constraints (e.g., 'varies' type for param_* is not a valid JSON Schema type). No error handling guidance, no output structure documentation, and no security considerations for executing arbitrary code. The server exposes code execution as a tool parameter without dry-run or confirmation patterns, creating irreversible risk.

• Document output schema for execute: return {success: boolean, output: string, error?: string, executionTime: number}. LLMs need to know what fields to expect.
• Replace param_* with explicit named parameters in dynamic_command_tools schema. Use proper JSON Schema types (string, integer, boolean, array, object) instead of 'varies'.
• Add error handling guidance to all tools: 'If the script fails, check syntax with execute({code: "...", language: "jxa", dryRun: true}) first.' Implement dry-run mode for execute.
• Document dynamic tool output structure: dynamic_list_tools returns [{_index: number, [property]: any}]; dynamic_get_tools returns {_index?: number, [property]: any}. Include this in descriptions.
• Add confirmation pattern for execute: require a confirm parameter or implement a two-step flow (preview → execute) to prevent accidental code execution.