Four tools with complete Zod schemas and descriptions, but significant quality gaps. Tool names lack action verbs (gemini_think, gemini_brainstorm, gemini_review, gemini_query are descriptive but not verb-noun patterns). Descriptions are adequate (100-150 chars) but lack WHEN/WHY guidance. Parameters have types and descriptions, but output schemas are completely undocumented, LLMs cannot plan downstream calls. No error handling guidance. No tool annotations (readOnlyHint, destructiveHint). All tools are READ_ONLY but this is not formally declared in the schema.

• Add explicit output schemas to each tool. E.g., gemini_think should document: {analysis: string, key_points: string[], recommendations: string[]}. This enables LLMs to extract and chain results.
• Rename tools to verb_noun pattern: 'analyze_problem' (gemini_think), 'generate_ideas' (gemini_brainstorm), 'evaluate_code' (gemini_review), 'query_gemini' (gemini_query). Clearer intent, better LLM disambiguation.
• Expand tool descriptions with WHEN/WHY guidance. E.g., 'Use analyze_problem for deep reasoning on complex technical or strategic questions. Use generate_ideas for creative brainstorming. Use evaluate_code for code review and architecture critique.'
• Add error handling to callGemini(). Catch API errors and return structured error responses with recovery hints: 'Gemini API failed: {error}. Retry with a simpler prompt or check API credentials.'
• Declare tool annotations in schema: add readOnlyHint: true to all tool definitions to formally signal safety properties to agents.