26 tools with READ_ONLY risk classification. All tools have names starting with action verbs (get_, search_, run_). Descriptions present but vary widely in quality (10-200 chars). Input schemas visible with type declarations, but many lack detailed parameter descriptions. Output schemas not documented. Error handling guidance absent. Tool composition is reasonable (single responsibility), but parameter constraints are minimal. No tool annotations (readOnlyHint, idempotentHint). Naming is clear but some tools are overly generic (e.g., 'kimi_datasource', 'tv_scan', 'tv_analyze' require action parameter to disambiguate). Parameter descriptions often missing or trivial.

• Document output schemas for all 26 tools. Include field names, types, and descriptions. Example: get_quote should return {symbol: string, price: number, currency: string, timestamp: ISO8601, source: string}.
• Replace free-form 'action' parameters with enums. For tv_scan, enumerate valid actions (e.g., 'top_gainers', 'top_losers', 'most_active', 'overbought', 'oversold'). For quant_backtest, document all valid actions in the schema, not just in the description.
• Expand 'source' parameter descriptions to list valid values per tool. Example: 'Data source: auto (auto-select by market), futu (Futu OpenD), yahoo (Yahoo Finance), fmp (Financial Modeling Prep), alphavantage (Alpha Vantage), eastmoney (East Money for CN A-shares).'
• Add error handling guidance to tool descriptions. Example for get_quote: 'If symbol not found, call search_symbols() to find the correct ticker. Returns error code 404 if source is unavailable.'
• Add tool annotations (readOnlyHint: true) to all tools in the tool definition. This makes safety properties machine-readable and helps agents reason about side effects.