Single tool 'web_search' has a basic schema with two parameters (query, max_results) but critical gaps undermine quality. Description is minimal (11 words, ~60 chars), below the 10 - 1024 char baseline and lacks WHEN/WHY context. Parameter 'max_results' is marked optional but has no type constraint (number range), no enum, no default. Query parameter lacks format guidance. No output schema documented, LLMs cannot plan downstream calls or extract structured data. No error handling guidance (e.g., what if Tavily API fails? Rate limited?). Tool does not follow verb_noun naming (web_search is acceptable but generic). No indication of pagination support, result limits, or field structure returned. Composition risk: if this is part of a larger agent system, downstream tools cannot chain on undocumented response fields.

• Expand description to 150 - 250 chars: 'Search the web using Tavily API. Returns up to 20 results with title, URL, and snippet. Use when you need current information not in training data. Supports natural language queries.'
• Add max_results constraint: type number, minimum 1, maximum 20, default 10. Document in description: 'Maximum results to return (1 - 20, default 10).'
• Document output schema: array of objects with fields [url (string), title (string), snippet (string), source (string)]. Include example.
• Add error handling: 'If Tavily API fails or rate-limits, wait 60 seconds and retry. If no results found, try a simpler query or different keywords.'
• Add query parameter guidance: 'Natural language search query (e.g., "latest AI breakthroughs 2026"). Avoid boolean operators; Tavily handles intent automatically.'