Speekify has 2 tools with complete input schemas and descriptions. Tool naming follows verb_noun convention (generate_wav, generation_defaults). Descriptions are present but brief (18-28 chars for tool names, ~100 chars for descriptions). The generate_wav tool has 12 parameters with types and defaults, but parameter descriptions are minimal (10-30 chars each), below the 72-char baseline. Output schema is documented in code (_serialize_generation returns 7 fields) but not exposed in tool definition. No error handling guidance, no recovery hints, no parameter constraints (e.g., voice enum, speed range). The generation_defaults tool is read-only and well-scoped but lacks output schema documentation.

• Expand parameter descriptions to 50-100 chars. Example: 'voice: Supertonic voice model. M1-M5 are male voices (M1=deep, M5=high); F1-F5 are female voices (F1=deep, F5=high). Default M1 is neutral male.'
• Convert voice and language_code to enum constraints in schema. Expose VOICE_NAMES and SUPPORTED_TTS_LANGUAGES as enums so LLM can pick valid values without guessing.
• Document output schema in tool definition: {output_path: string (absolute file path), output_uri: string (file:// URI), duration_seconds: float, batch_count: int, title: string, text_length: int, warnings: array of strings, log_path: string}.
• Add error handling: catch file-not-found, invalid URL, synthesis failures, and return structured errors with recovery hints. E.g., 'Invalid voice M6: must be one of M1-M5, F1-F5. Retry with a valid voice.'
• Add parameter constraints to schema: speed range [0.5, 2.0], steps range [10, 100], max_chunk_length > 0 if provided. Document these in descriptions.