voice-batch-runner

Voice Batch Runner

Use When

• Persona, concept, and script inputs already exist and the next step is hosted voice design, cloned voice take generation, or polling.
• The voice should remain a durable persona asset across scripts, not a one-off audio byproduct.

Do Not Use When

• The task belongs to ideation, QA, or another released skill listed in the handoff section.
• Required inputs are missing and guessing would change the result.
• Voice strategy, task class, translation policy, or lip-sync intent is still unresolved. Use audio-generation first.

Execution Boundary

• This runner validates and executes normalized voice requests. It must not make creative strategy, voice-policy, or reference-policy decisions.
• Separate the workflow into voice profile, optional voice identity, and concrete voice take. The script text can change; persona voice continuity should not.
• Voice design is for an initial persona-aligned sound from text, voice_description, and language.
• Voice clone is for new script takes when approved reference audio and an optional reference_text should preserve timbre and speaking style.

Source And Path

• Ground requests in the active project persona registry, voice baseline, script text, and video purpose/lane.
• Use the active project/client folder first; do not assume one client directory is the source base for all voice work.
• Keep internal request/response/run state under .postplus when it is not the user-facing handoff. Keep final audio and review files in the active voice asset folder, or state the chosen workspace path.

Request Boundary

• Voice design synthesizes a persona voice from spoken text, a free-text voice_description, and an optional language (defaults to auto).
• Voice clone reproduces an approved voice from spoken text, an audio reference, an optional reference_text transcript, and optional language. Upload a local reference with postplus media-file upload --skill voice-batch-runner --input-file <file> --mime <audio/mpeg|audio/wav> --output <upload.json>, then pass the persistent output.mediaReference as --audio; a remote HTTPS URL also works, but must remain reachable through submission.
• Exact field names, requiredness, and defaults are discovered from postplus media schema --json and the generated example below; do not hard-code a private request envelope here.

Review And Handoff

• Before generation, verify persona registry, voice baseline, script stability, route (voice_design or voice_clone_take), source basis, and output path.
• After generation, review realism, persona fit, pacing, ad-like delivery, reuse potential, and for cloned output, timbre/accent drift from the reference.
• If pending, return the manifest path, the output.data.id generation handle, and the poll command postplus media poll --handle <output.data.id>. The poll waits in-command (up to 45s per invocation, checking every 8s); rerun it while pending instead of writing a tighter retry loop.
• Save a finished take with postplus media-file download --url <fresh-output-url> --output-file <path> (or --reference <postplus-media://...> for hosted storage); provider output URLs are temporary, so download while fresh.

Stop Conditions

• Stop when required user intent, source evidence, or owned input artifacts are missing and guessing would change the result.
• If an owned CLI or script command fails, report the exact error and stop. Do not bypass the failure with metadata-only answers, readiness probing, local payload rewrites, fallback providers, or unpublished tools.
• Batch isolation: when producing a batch of independent items, a per-item provider content/safety rejection is isolated to that item. It is identified only by the typed code postplus_cli_hosted_media_content_policy_blocked, never by provider prose, and it surfaces at either boundary: a failed postplus media create whose typed error code is that code, or a submitted run whose poll result carries output.data.status: failed and output.data.error.code set to that code. On either, record which item was blocked and its exact reason, skip it, and continue submitting and polling the remaining items, then report the incomplete set at the end. Do not retry, soften, or re-submit the blocked item — that is a forbidden payload rewrite. Every other failure (a failed owned CLI/script command whose typed code is not that content-policy code, or a run whose error.code is not that content-policy code — auth, transport, quota, malformed request, provider outage) is systemic: stop per the rule above.

Public Command Boundary

• Choose the smallest matching command or workflow from the user input and run it directly.
• Readiness diagnostics: postplus doctor --skill voice-batch-runner.
• Poll a pending voice take: postplus media poll --handle <output.data.id> (waits in-command up to 45s per invocation; rerun while pending).
• If an owned CLI or script command fails, report the exact error and stop. Do not bypass the failure with metadata-only answers, readiness probing, local payload rewrites, fallback providers, or unpublished tools.
• Use postplus media schema --json only when constructing or repairing an unknown request shape.
• Run the hosted submit with the generated command below; do not call provider APIs directly.

<!-- BEGIN GENERATED EXECUTION EXAMPLE -->

postplus media create voice-design \
  --text <text> \
  --voice-description <voice-description> \
  --output <result.json>

<!-- END GENERATED EXECUTION EXAMPLE -->

• If the CLI returns a quote-confirmation challenge, run postplus quote confirm --json --challenge-file <challenge.json> and retry with the returned token.