12 tools with basic descriptions and partial schemas. Most tools have 1-2 parameter descriptions but lack comprehensive output schema documentation. Tool names follow verb_noun convention (navigate, get_page_content, query_selector, execute_script, click, type_text, wait_for_selector, screenshot). Descriptions are present but brief (10-50 chars), below the 50-200 char LLM-optimized baseline. Input schemas visible for all tools with type declarations, but output schemas are not documented. No error handling guidance, no recovery hints, no parameter constraints (enums, ranges). Security concern: execute_script accepts arbitrary JavaScript with no validation or sandboxing hints.

• Document output schemas for all 12 tools. Specify what fields each returns (e.g., navigate returns {success: bool, url: string, loadTime: number}). This enables LLMs to chain tools and extract data.
• Expand descriptions to 50-200 chars. Add WHEN to use (vs similar tools), prerequisites, and what it returns. Example: 'Get the current page's rendered text content (use instead of get_page_content for plain text). Returns the full page text or empty string if no content.'
• Add enum constraints to navigate's waitUntil parameter: ['load', 'domcontentloaded', 'networkidle0', 'networkidle2']. Add min/max to wait_for_selector's timeout: 100-300000 ms.
• Add error recovery hints to descriptions. E.g., query_selector: 'If selector not found, try wait_for_selector() first to ensure element is in DOM. Returns null if not found.'
• Restrict execute_script with explicit safety guidance: document timeout (e.g., 30s), return type (JSON-serializable), and forbidden operations (e.g., 'Cannot access window.location or make network requests'). Consider a separate tool for safe, pre-approved scripts.