sumi-voicebox — Claude Code skill: what it does, how to install
sumi-voicebox is a Claude Code skill that configures local Voicebox narration for Sumi videos. It triggers when choosing on-device voice generation without cloud API costs.
MIT ★ 110 updated 2026-11-07
npx skills add beyondtahir/beyondsumi --skill sumi-voicebox -a claude-codeWhat sumi-voicebox does
The skill configures offline speech generation using the upstream jamiepine/voicebox project without hosted character fees. It reads system capacity and loopback status through python3 scripts/probe.py without triggering background downloads.
Model downloads and runtime components require explicit scope, showing exact source targets, disk footprint, and install paths. Synthesis operations query the local /openapi.json endpoint, tracking job IDs to avoid duplicate tasks. The skill keeps recordings private and prohibits exposing the backend to the public internet.
What it helps with
- check local hardware capacity with
python3 scripts/probe.py - configure on-device Voicebox narration without character billing
- verify model downloads with exact size estimates and targets
- track local generation jobs using the
/openapi.jsonendpoint
Example prompts for the agent
- Set up local Voicebox narration for my story video project
- Run probe.py to verify hardware status before downloading models
- Generate voice narration from approved text using local openapi.json
How to install sumi-voicebox
1Skills CLI (Vercel)
npx skills add beyondtahir/beyondsumi --skill sumi-voicebox -a claude-code2Manually: put the files in place
~/.claude/skills/sumi-voicebox/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/sumi-voicebox/
1Skills CLI (Vercel)
npx skills add beyondtahir/beyondsumi --skill sumi-voicebox -a codex2Manually: put the files in place
.agents/skills/sumi-voicebox/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add beyondtahir/beyondsumi --skill sumi-voicebox -a cursor2Manually: put the files in place
.cursor/skills/sumi-voicebox/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install beyondtahir/beyondsumi skills/sumi-voicebox2Manually: put the files in place
.github/skills/sumi-voicebox/SKILL.md1Skills CLI (Vercel)
npx skills add beyondtahir/beyondsumi --skill sumi-voicebox -a gemini-cli2Platform CLI
gemini skills install https://github.com/beyondtahir/beyondsumi --path skills/sumi-voiceboxManually: put the files in place
.devin/skills/sumi-voicebox/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about sumi-voicebox
How to install sumi-voicebox in Claude Code?
Run npx skills add beyondtahir/beyondsumi --skill sumi-voicebox -a claude-code.
What does sumi-voicebox do?
It guides local Voicebox narration setup, monitors hardware limits via probe scripts, and tracks generation jobs through the local API.
What is required to run sumi-voicebox?
The setup relies on python3, a local Voicebox release, and access to the local /openapi.json interface.
Source text
SKILL.md · MIT
--- name: sumi-voicebox description: Guide optional local Voicebox narration setup, hardware/model-size checks, consented voice profiles and export for Sumi or other story videos; downloads and generation require the user's selected scope. --- # Voicebox narration setup Use [the verified setup guide](references/setup.md) when the user selects free local narration. Explain app versus selected speech model, hardware/storage and no per-character hosted API bill. Run `python3 scripts/probe.py` for read-only loopback status; no install/download/generation is triggered. Use the exact upstream [Voicebox repository](https://github.com/jamiepine/voicebox), official releases and current model/license information. Do not vendor models or clone the full repository just to install a desktop release. Reuse a compatible existing app/model. Ask before missing third-party app/runtime/model downloads, with exact selected items, source, size estimate and location. Do not install every engine. Offer a licensed preset, user's own/consented reference, supplied narration or no narration. A distinctive voice can aid identity but does not guarantee platform monetization. Check the installed local `/openapi.json` and model status before API mutations. Generate approved text once with the verified model/profile and retain job ID/take; inspect an uncertain existing job before retrying. Use the normal Generate/Export interface if this version's API differs. Never expose the backend publicly or silently use cloud services. Save non-secret preferences only; recordings and models remain private.
© beyondtahir · MIT · beyondtahir/beyondsumi