Server has 5 tools with basic schemas and descriptions in Chinese. All tools have input schemas with type definitions and descriptions present. However, descriptions are domain-specific (Chinese language) and lack LLM-optimization guidance (e.g., when to call vs alternatives). Parameter descriptions exist but are minimal. No output schemas documented. Error handling returns text but lacks recovery guidance or error classification. Tools mix read-only and write operations without clear permission semantics. No tool annotations (readOnlyHint/destructiveHint). Naming follows verb_noun pattern (scan_skills, suggest_additions, list_known, clear_cache, export_report) which is good, but descriptions don't explain when to use one tool vs another.

• Add explicit output schemas to all tools. Document the structure of scan_skills response (e.g., {skills: [{name, category, status}], coverage: {percent, gaps: []}}, analysis: {...}). This enables LLMs to extract and chain results.
• Add tool annotations: mark clear_cache and export_report with destructiveHint=true; mark scan_skills, suggest_additions, list_known with readOnlyHint=true. This signals to agents which tools have side effects.
• Expand descriptions to 100-150 chars each, explaining WHEN to call each tool. E.g., 'scan_skills: Discover installed Agent Skills and identify gaps. Call this first to understand current coverage. Returns a report with skill names, categories, and coverage percentage.' vs current '自动扫描...'
• Enhance error messages with recovery guidance. Instead of '❌ 错误: File not found', return '❌ Export failed: output directory does not exist. Create the directory or provide a valid path (e.g., ./reports/).'
• Document parameter constraints in descriptions. For exportPath, add '(must be a writable directory path, relative or absolute)'. For exportFormat, explain when to use each: 'text for human reading, json for programmatic parsing, markdown for documentation'.