Scoring was not performed

• Document output schemas for all 4 tools. Example for search_knowledge_base: {type: 'object', properties: {results: {type: 'array', items: {type: 'object', properties: {item_id: string, title: string, content: string, risk_type: string, relevance_score: number, chunk_index: number}}}, total_count: number, limit: number}
• Add 'TopK: 5' or similar pagination parameter to search_knowledge_base and document the limit in the tool description: 'Returns up to 5 results. Use the limit parameter to adjust (1 - 20).'
• Provide English descriptions alongside Chinese descriptions to improve accessibility. At minimum, add English short descriptions in the schema.
• For read_skill, add error recovery guidance: 'If skill_name is not found, call list_skills first to see available skills.'
• For search_knowledge_base 'risk_type' parameter, add a constraint note: 'Must match one of the values returned by list_knowledge_risk_types. Leave empty to search all types.'
• Add error classification to all tools: when an error occurs, return a structured response indicating if it's retryable (e.g., search timeout) or user-fixable (e.g., invalid skill_name).
• For list_knowledge_risk_types and list_skills, add output schema documenting the structure of the returned text (e.g., list format, field names if structured) and consider returning JSON arrays instead of formatted text for machine parsing.
• Add 'short_description' field to each tool (1 - 20 char summary) to improve LLM tool selection speed.