Three tools with basic descriptions and parameter schemas, but significant gaps in schema completeness, error guidance, and output documentation. All tools are READ_ONLY (good), but descriptions lack actionable context for LLM selection. Parameters have types and defaults but lack detailed constraints. No documented output schemas. Error handling returns plain strings without recovery guidance.

• Add explicit output schema documentation. Define financial_search to return {results: [{content: string, source: string, score: float}], total: int}. Define knowledge_graph_query to return {triples: [{head: string, relation: string, tail: string}], entity: string}. Define financial_analysis to return {analysis: string, sources: [string], confidence: float}.
• Enhance error messages with recovery guidance. Replace 'Retrieval failed: {e}' with 'No documents matched query. Try: (1) simplify query, (2) use search_financial_search with broader terms, (3) check if data is loaded.' Categorize errors as retryable vs user-fixable.
• Add parameter constraints to descriptions. For top_k: 'Integer 1-100, default 5. Limits results to prevent context overflow.' For relation: 'Optional filter; valid types: 同比增长, 环比增长, 同比下降 (see knowledge_graph_query docs for full list).' For query: 'Chinese or English text, 1-500 chars. Examples: 贵州茅台2024年营收, Moutai revenue 2024.'
• Clarify tool selection logic in descriptions. Add to financial_search: 'Use for document retrieval when you need specific data points or sources.' Add to knowledge_graph_query: 'Use for entity relationships and comparative analysis.' Add to financial_analysis: 'Use for synthesis, trend analysis, or multi-entity comparisons requiring LLM reasoning.'
• Enforce and document result limits. Cap knowledge_graph_query triples at 50 and financial_search results at 20. Add to descriptions: 'Returns up to N results; use pagination (offset/limit) for larger datasets.' Implement offset/limit parameters if not present.