Two tools with basic schemas and descriptions, but significant gaps in parameter documentation and error handling. voice_speak has a clear purpose but lacks output schema documentation. voice_toggle uses an enum for action parameter (good), but both tools lack descriptions of what they return and how to handle failures. No error recovery guidance. Descriptions are adequate (100-150 chars) but parameter-level documentation is sparse. No tool annotations (readOnlyHint/destructiveHint). Naming is verb-forward (voice_speak, voice_toggle) which is correct, but descriptions don't explain prerequisites, dependencies, or failure modes.

• Add explicit output schema to voice_speak: {success: boolean, message: string, duration_ms?: number, tts_engine_used: 'local'|'gemini'|'fallback'}. Document what each field means.
• Add explicit output schema to voice_toggle: {action: string, listening: boolean, response_enabled: boolean, available_actions: string[], error?: string}. Clarify which fields are present for each action.
• Expand parameter descriptions: 'interrupt (boolean): If true, stop any in-progress playback before speaking. If no playback is active, this is a no-op. Default: false.' Add similar clarity to voice_toggle action enum values.
• Add error handling guidance to both tool descriptions. E.g., 'If TTS fails, voice_speak returns {success: false, message: "TTS unavailable: <reason>"}. Retry with a shorter text or check audio device.' This guides LLM recovery.
• Add tool annotations to schema: voice_speak should include 'readOnlyHint: true' in metadata; voice_toggle should include 'destructiveHint: true' for 'stop' action. Helps LLMs reason about idempotency and retry safety.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry