Class Duty Voice Reminders

What problem it solves

Teachers may already have a finished class-duty schedule, but still need to turn it into labeled spoken reminders for class or dorm channels. A naive prompt that says “generate voice from the table” can create two problems: it may add missing names, punishments, scores, or duties that are not in the table, and it may mix free planning, voice cloning, and paid narration into one unclear step. This skill separates reading, sentence splitting, payment confirmation, generation, review, and recovery into a controlled workflow.

How it works

• Hard inputs first: it only uses the teacher's existing duty table, pronunciation list for names/places/custom terms, target count, and voice rights. The default count is 10 items, within the 8-20 range.
• Free slot table first: before any paid call, it builds a labeled spoken-sentence list, such as Monday, sweeping, blackboard, watering, and notes whether each slot uses a catalog voice or a cloned voice.
• Staged confirmation: cloning and narration are two separate paid cards. Cloning requires an authorized sample first; narration requires reading beatra.models.list, pricing, and per-slot submission via beatra.speech.synthesize.
• Submit and recover: all remote calls go through the bundled scripts/mcp_client.py, sentences are split, each input stays under 50,000 characters, and results are polled with beatra.tasks.get while checking audio MIME, duration, size, and actual billing.

Boundaries and cautions

The skill does not create new duty assignments or invent names, tasks, and penalties that are absent from the table. If names exist but the pronunciation list is empty, it should stop and collect pronunciations. If the wallet is low, it reports the official guidance without retrying paid work. It is best treated as a controlled schedule-to-audio workflow for teachers who already have a clear duty table and need a batch of short spoken reminders.