Problem

Voice-only channels (e.g., `voice-call` extension) are unusable without STT (inbound voice → text) and TTS (outbound text → voice). Today:

• No `voiceOnly` capability field exists on `ChannelCapabilities`
• No validation during onboarding or channel setup that TTS/STT credentials are configured
• Errors only surface lazily at runtime when an agent tries to use the channel
• STT config is plugin-specific (voice-call only), not global
• TTS has a free fallback (Edge) but STT has none — voice channels silently break without STT credentials

After #424 (STT as middleware) and #402/#403 (TTS auth unification), both STT and TTS credentials live in the global auth profile store. Voice-only channels should validate that the required auth profiles exist.

Scope

1. Add `voiceOnly` to `ChannelCapabilities`

`src/channels/plugins/types.core.ts`:
```typescript
export type ChannelCapabilities = {
chatTypes: Array<ChatType | "thread">;
voiceOnly?: boolean; // channel only supports voice I/O
// ... existing fields
};
```

`voice-call` extension sets `voiceOnly: true` in its capabilities.

2. Validation at channel enable time

When a channel with `voiceOnly: true` is enabled (onboarding or `remoteclaw channels add`):

• Check that a TTS provider is configured with valid auth credentials
• Check that an STT provider is configured with valid auth credentials
• If either is missing: warn the user, explain what's needed, optionally block enabling

3. Runtime validation

Before accepting messages for a voice-only channel:

• Verify TTS is available (can resolve API key for configured TTS provider)
• Verify STT is available (can resolve API key for configured STT provider)
• If unavailable: return actionable error to the channel ("Voice channel requires TTS/STT configuration")

4. Health check integration

`remoteclaw doctor` should flag:

• Voice-only channels enabled without TTS credentials
• Voice-only channels enabled without STT credentials

Design notes

• Edge TTS is free (no API key) — counts as valid TTS for this validation
• STT has no free fallback — always requires credentials
• Validation checks auth profiles via `resolveApiKeyForProvider` from `src/auth/`
• This does NOT add TTS/STT prompting to onboarding (that's a separate concern per feat: unified auth + multimodal AgentRuntime — implementation plan #415 Phase 5/6 notes)

Depends on

• refactor(media): extract STT from media-understanding into src/stt/ #424 — STT extraction to `src/stt/` (STT as middleware, not plugin-specific)
• fix(tts): resolve TTS API keys through auth profile store #402 — TTS auth integration (TTS credentials from auth profiles)
• feat(auth): add ElevenLabs provider to auth profile system #403 — ElevenLabs auth provider

Related

• feat: unified auth + multimodal AgentRuntime — implementation plan #415 — implementation plan (parent)
• feat(middleware): communicate multimodal limitations to users when runtime can't handle media #400 — multimodal limitation notices (related user feedback)
• `extensions/voice-call/` — primary voice-only channel