Four tools with complete input schemas and clear descriptions. Naming follows verb_noun pattern (generate_image, search_images, get_image, list_categories). Descriptions are well-written (100-200 chars, within baseline 194 avg). All parameters have type definitions and descriptions. However, output schemas are not documented, LLMs cannot predict response structure for chaining. Error handling guidance is absent. Parameters lack enums where appropriate (e.g., size, orientation, lang should be constrained). No mention of pagination limits for search_images despite potentially returning many results.

• Document output schemas for all four tools. Example for search_images: 'Returns {results: [{id: number, path: string, subject: string, prompt: string, rating: number}], total: number, top_match: {id, path, data_uri}}'.
• Convert 'size' parameter to enum: ['square', 'landscape', 'portrait', 'hd', 'fhd', 'qhd', 'uhd'] plus a pattern for custom WxH. Same for 'lang' and 'orientation'.
• Add error recovery hints to descriptions: generate_image → 'If text is detected, try adding "no text" to the prompt or set no_text=true'; search_images → 'If no results, try broader keywords or remove category filter'.
• Document pagination for search_images: 'Returns up to 5 results by default (limit param). Use limit and offset for pagination. Always returns total count.'
• Clarify get_image parameter relationship: 'Provide either id (from search_images) or path (absolute vault path). If both provided, id takes precedence.'