Server has 5 tools with complete input schemas and descriptions. Naming follows verb_noun convention (skills_catalog_*). Descriptions are present (avg 140 chars) but lack actionable context for LLM selection. Parameters are typed but many lack constraints (enums, ranges). Output schemas are not documented, LLMs cannot predict response structure. Error handling is minimal; migrate tool has partial error detection but no recovery guidance. No tool annotations (readOnlyHint/destructiveHint). Composition is sound: each tool has one responsibility, but chaining IDs are not guaranteed in responses.

• Add output schema documentation to each tool. Define response structure: e.g. skills_catalog_find returns {results: [{id, name, domain, relevance_score, triggers}], total_count, next_cursor}. Include in tool description or as a separate schema field.
• Add tool annotations: mark skills_catalog_find, skills_catalog_list_domains, skills_catalog_validate as readOnlyHint=true; mark skills_catalog_generate and skills_catalog_migrate as destructiveHint=true.
• Enhance descriptions with WHEN/WHY context. E.g. 'Search for skills by trigger phrase, description, or domain. Call this first to discover available skills before generating new ones. Returns ranked results with relevance scores.'
• Add min/max constraints to numeric parameters. E.g. limit: {type: 'number', minimum: 1, maximum: 100, default: 5}. Add enum constraints to mode parameter in migrate (already present) and domain in find.
• Implement structured error responses with recovery guidance. E.g. {error: 'Skill not found', code: 'NOT_FOUND', retryable: false, suggestion: 'Call skills_catalog_find to search for available skills.'}. Categorize errors in catch block.