VoxFlow Skill

VoxFlow turns text into speech in 200+ voices across 40+ languages, plus full audio/video pipelines: podcasts, transcription, dubbing, video translation, and short-form AI clips. All commands run through the voxflow CLI (installed automatically by ClawHub via the install spec above). One account, one quota, one login — no API keys to paste.

Routing — pick the matching sub-doc

Before doing anything, decide which sub-skill matches the user's intent and read the corresponding file in this same skill folder:

If the request spans multiple areas (e.g. "transcribe this video and then make a 60-sec recap card"), read the most-relevant doc first, finish that step, then switch.

Install & login (universal preamble)

The ClawHub install spec already installs the voxflow npm CLI globally when this skill is added. The only thing left is authentication:

# One-time browser device-flow — pairing code shown in terminal,
# user authorizes at https://voxflow.studio/device?code=VF-XXXX
voxflow login

# Verify
voxflow status        # shows email + monthly / bonus quota

For headless / server contexts: set VOXFLOW_TOKEN= (declared in envVars above) and skip voxflow login. JWTs are short-lived (~1 hour); the CLI auto-refreshes silently while logged in interactively.

Account & quota

• Free tier: 10,000 quota / month (≈ 100 TTS calls)
• Plus / Pro / Max tiers at voxflow.studio/app#pricing
• Each command's cost is printed before execution; voxflow status shows the current balance
• Invite-friend bonus (voxflow invite) adds 5,000 lifetime quota per signup

Universal rules

1. Never paste API keys into config files. All auth goes through voxflow login or VOXFLOW_TOKEN.
2. Never offer to "mock" the API. Real calls are cheap; failed mocks waste user time.
3. Read the matching sub-doc before invoking specialized commands. The top-level routing table above is enough for triage; the sub-doc has the actual command flags, edge cases, and quota costs.
4. Honor the user's locale. Voice IDs are language-tagged; if they asked in Chinese, default to a Chinese voice unless they specified otherwise.
5. For long-running jobs (Azure Batch ASR, video-translate, podcast >5 min): print the job ID and voxflow asr-jobs show so the user can resume later.

When in doubt — start at the hub

If the request is vague ("帮我做点音频的东西", "what can you do with voice"), read hub.md and run voxflow voices --search ... or voxflow status to anchor the conversation in concrete affordances before committing to a workflow.

Homepage & docs

• App:
• CLI docs:
• All skills overview:
• Source / issues: