Three well-defined tools with complete JSON schemas, clear descriptions, and proper enum constraints. All tools are read-only and stateless. Naming follows verb_noun pattern (generate_ipsum, list_themes, theme_lore). Descriptions are concise (50 - 150 chars) and explain purpose clearly. Parameters are typed with enums for theme/units/format. However, output schemas are not documented in the tool definitions, LLMs cannot see what fields to expect in responses. Error handling is minimal; no recovery guidance provided. Missing parameter descriptions for some fields (e.g., 'seed' lacks detail on reproducibility guarantees).

• Document output schemas for all three tools. Example for generate_ipsum: 'Returns {text: string, seed: string, theme: string, units: string, count: number}'. This enables LLMs to chain calls and extract the right fields.
• Expand 'temperature' description: 'Blend dial 0 - 1. At 0, output is pure Latin (Cicero). At 1, output is pure voice (e.g., Yass Kween). Intermediate values blend both. Defaults to the theme's natural level.'
• Add error recovery hints: 'If theme is invalid, call list_themes() to see available options. If seed produces unexpected output, try a different seed or increase temperature.'
• Clarify 'seed' parameter: 'A string seed for reproducible output. Same seed + options always yield identical text. Accepts any string; no length limit. Useful for testing or generating consistent placeholder text across multiple calls.'
• For list_themes, document that it returns an array of {id: string, name: string, description: string} objects, enabling agents to extract theme IDs for downstream generate_ipsum calls.