Class Duty Voice

---
name: "class-duty-voice"
description: "Turn a written class duty roster into one class duty voice clip per labeled cue. This duty roster voice studio records each class duty audio and roster reminder from the list the teacher already wrote, then delivers 8 to 20 class duty voice pack files. Use it for duty clip packs that keep one roster item on each clip."
---

# Class Duty Voice Pack

Turn a written class duty roster into one spoken clip per labeled cue.
Deliver 8 to 20 reads. Cover weekday names and the listed chores first.

## Scope and adjacent routes

Use this Skill when a teacher needs labeled class-duty reminder clips
from the roster they already wrote.

Route generic narration to `voiceover-narration-studio`. Route a
homeroom weekly list to `homeroom-week-voice`. Route holiday homework
to `holiday-homework-voice`. Route a census schedule to
`census-notice-voice`.

Read only the supplied duty roster. Do not invent who skipped duty, a
punishment, a grade, or a missing name.

## Collect the roster

Hard inputs are:

- the class duty roster the teacher already wrote;
- a pronunciation table whenever names, places, or coined words
  appear;
- how many clips the pack should contain, or permission to use the
  default of 10;
- likeness and voice rights when a cloned staff voice will appear.

Reuse already-known language and a frozen `voice_id`. Ask only for a
missing hard input. A count outside 8 to 20 is still doable: confirm
that pack size and its live cost.

Do not invent who skipped duty, a punishment, a grade, or a missing
name. File access is not consent.

If names exist and the pronunciation table is empty, stop and collect
the readings.

## Plan the free slot list

Write a labeled class-duty voice list before any paid clone or
speech. Default ten slots unless the teacher names another count in
8 to 20: Monday, Tuesday, Wednesday, Thursday, Friday, weekend,
broom, chalkboard, watering, and extra. Each slot records the spoken
line from the written roster and whether it uses a catalog voice or a
clone. Omit a default role whose line is missing on the roster.

That list is the free visible result. Planning is not approval.

Safe defaults:

- one `beatra.speech.synthesize` call per slot;
- `model: "auto"` only when every live speech card supports the
  language;
- `format: "mp3"`; `speed: 1.0`;
- one brand voice for the whole pack.

Keep each submitted `input` at or below 50,000 characters. Split on
sentence boundaries. Write short spoken sentences from the roster.

Inspect an authorized clone sample when clone is requested. For a
local file, upload only through the bundled client after inspection
(`scripts/mcp_client.py` / `beatra.assets.upload`). Keep the returned
artifact id. Never pass a local path to `beatra.voices.clone` or
`beatra.speech.synthesize`.

## Confirm clone, then speech

Clone and speech are separate paid stages. Each stage gets its own
six-field card and its own opaque `client_request_id`.

If the teacher wants a cloned staff voice, inspect an authorized
sample, read the live `voice_clone` card, and wait on the clone card
before `beatra.voices.clone`. A found file is not clone consent. Show
the clone card and wait:

1. Work — one authorized voice sample (`beatra.voices.clone`).
2. Credits — the live `voice_clone` price just read. Do not reuse a
   remembered number.
3. Count — one paid clone call for this sample.
4. Identity — one new opaque `client_request_id`.
5. If we stop here — the labeled slot list remains usable.
6. If the balance is insufficient — relay the official message and
   its top-up URL exactly. Translate the prose; keep the URL. Do not
   retry until the teacher says they have topped up.

Before speech, call `beatra.models.list` for `text_to_speech`:

```json
{"capability": "text_to_speech"}
```

Use `beatra.voices.list` when a catalog voice still needs a choice.
Never put a display name in `voice`. Show the speech card and wait:

1. Work — one spoken roster cue per named slot
   (`beatra.speech.synthesize`).
2. Credits — the live `text_to_speech` price just read, times the
   slot count. Do not reuse a remembered number.
3. Count — one paid speech call per slot.
4. Identity — one new opaque `client_request_id` per slot.
5. If we stop here — the labeled slot list remains usable.
6. If the balance is insufficient — relay the official message and
   its top-up URL exactly. Translate the prose; keep the URL. Do not
   retry until the teacher says they have topped up.

Submit each speech slot once through bundled `scripts/mcp_client.py`.
Poll `beatra.tasks.get`. Read actual audio MIME, duration, and size,
and `billing.net_charged_credits`. Do not promise the prepaid
estimate is the final charge. A script preview is not the audio
review.

## Review, deliver, and recover

Review that every clip uses the same voice and that names match the
pronunciation table and the written duty roster. Report only
what the host can actually hear. Do not invent who skipped duty.

After a returned `task_id`, poll that task. If the create response is
lost, search with `beatra.tasks.list` and verify with
`beatra.tasks.get` before replay. Reuse an ID only with
byte-identical arguments. A changed line, voice, or speed is a new
card and a new ID. Cancel only when the teacher asks.

## Execution

Invoke every remote Beatra operation only through this package's bundled
`scripts/mcp_client.py`. Put the MCP tool name after `call` and send one
JSON object on standard input.

```bash
python3 scripts/mcp_client.py call beatra.models.list
```

```json
{"capability": "text_to_speech"}
```

```text
printf '%s' '{"input":"<the written line for this slot>","voice":"voice_...","format":"mp3","client_request_id":"opaque-duty-speech-01"}' | python3 scripts/mcp_client.py call beatra.speech.synthesize
```

Do not configure or call a host Beatra Connector, and do not use
REST/OpenAPI as a fallback.

If the tool returns insufficient credits: give the user the tool message in
full. Do not change the URL. You may translate the other words. Do not replace
it with your own top-up copy. Point the user to pay using that message.

## Account balance

When the user asks how many credits remain or whether a live estimate fits,
call `beatra.wallet.get`. When they ask what was charged, call
`beatra.wallet.ledger`. Both are read-only. Do not invent an account-balance or
top-up tool. Do not make `wallet.get` a required step before every paid submit.

When a model card comes back carrying a `top_up` block, relay its tiers as the
card lists them and in that order. Do not rank them, do not talk one down, and
do not pick one for the user. Which tier suits them is their call, made on
the wallet page with the whole list in front of them. Never quote a tier from
memory.

## References by task

- For slot lists, payloads, and recovery, read
  [Class duty voice workflow](references/workflow.md).
- For authorization and the non-billable registration step, read
  [installation and authentication](references/installation-and-auth.md)
  and [installation registration](references/installation-registration.md).
- For shared task, billing, and connection details, read [tasks and
  results](references/tasks-and-results.md), [billing, errors, and
  recovery](references/billing-errors-and-recovery.md), and [Bundled MCP
  Client diagnostics](references/mcp-connection.md).
- For update guarantees and controls, read [automatic updates and
  safety](references/automatic-updates-and-safety.md). For removal, read
  [uninstall and disconnect](references/uninstall-and-disconnect.md).

## Runtime and safe automatic updates

The bundled client silently checks for a newer release at most once
every 24 hours per installation. When a newer version is available, it
installs automatically without separate confirmation. It downloads only
from the fixed official Beatra discovery and immutable CDN paths for
this package, channel, and locale, verifies discovery data, archive,
manifest, and every packaged file, and replaces only package-owned
files.

Update checks, downloads, verification, replacement, rollback, and
recovery fail open: the current installation remains usable and the
original command continues. An update failure never authorizes retrying
a paid clone or speech request. The setting persists for this
installation. See
[automatic updates and safety](references/automatic-updates-and-safety.md).

```text
python3 scripts/mcp_client.py update --auto off
python3 scripts/mcp_client.py update --auto on
python3 scripts/mcp_client.py update --check
```