fix(telegram): deliver TTS as additive final audio

[artemgetmann]

Review Fast Path

• User path fixed: Telegram /tts on final replies now deliver durable final text first, then a separate audio/voice message below it.
• Proof: targeted tests, lint/build, isolated live Telegram baseline pass, and live TTS scenario pass on runtime commit 1952d587af.
• Shared-state footgun removed: TTS no longer mutates a final text payload into a Telegram audio caption or delays final text delivery.
• Still hurts: progress-plus-tts live scenario still needs a reliable prompt/harness trigger for a separate progress event; TTS shape passed, progress emission did not.

Why This Matters

• Final text must remain visible, durable, and sent once.
• TTS should be additive: useful when it works, harmless when it fails.
• Telegram progress cleanup from fix(telegram): coalesce progress into one transient bubble #871 remains the baseline and must not regress.

Scope Boundary

• Changed final-mode TTS delivery for generic and ACP dispatch paths.
• Added a media-only TTS supplement helper and tests.
• Updated Telegram live harness readback to expose media_type and verify separate audio after final text.
• Did not restart or touch the shared main Telegram runtime.
• Did not broaden TTS provider/config behavior.

Verification

• pnpm test -- src/commands/telegram.test.ts src/cli/telegram-cli.test.ts src/auto-reply/reply/dispatch-from-config.test.ts src/auto-reply/reply/dispatch-acp-delivery.test.ts src/auto-reply/reply/dispatch-acp.test.ts src/tts/tts.test.ts pass: 136 tests.
• pnpm test -- extensions/telegram/src/progress-controller.test.ts extensions/telegram/src/bot-message-dispatch.api-sequence.test.ts extensions/telegram/src/bot-message-dispatch.high-route-api-sequence.test.ts src/agents/pi-embedded-subscribe.reply-tags.test.ts src/agents/pi-embedded-subscribe.handlers.tools.test.ts src/auto-reply/reply/agent-runner-payloads.test.ts src/auto-reply/reply/dispatch-from-config.test.ts src/auto-reply/reply/dispatch-acp-delivery.test.ts src/auto-reply/reply/dispatch-acp.test.ts src/tts/tts.test.ts src/commands/telegram.test.ts src/cli/telegram-cli.test.ts pass: 187 tests.
• pnpm lint pass.
• pnpm build pass.
• git diff --check pass.

Live Telegram Proof

• Isolated runtime ensure:
  • branch=codex/telegram-tts-additive-20260607
  • runtime_worktree=/Users/user/Programming_Projects/openclaw/.worktrees/telegram-tts-additive-20260607
  • runtime_commit=1952d587afbe07e1569064ab80fc9a697089f912
  • current_lane_bot=@Artem_jarvis_exec_bot
• Baseline pass:
  • pnpm openclaw:local telegram smoke baseline --json
  • artifact: .artifacts/telegram-smoke/20260607T110911Z-baseline-771cd206.json
  • reply: Baseline ping received and confirmed.
• TTS additive pass:
  • pnpm openclaw:local telegram scenario tts-final-caption --json
  • artifact: .artifacts/telegram-smoke/20260607T111615Z-tts-final-caption-d23788d0.json
  • final text message: 52365
  • separate voice/audio message: 52366
  • empty_voice_only_final_detected=false
  • mergeReadiness=sufficient
• Reset baseline:
  • /tts off sent after live checks
  • bot confirmed: 🔇 TTS disabled.

Known Gap

• progress-plus-tts on this isolated lane produced final text plus separate audio, but the model placed the progress note in the final text instead of emitting a separate progress event.
• Latest failing artifacts:
  • .artifacts/telegram-smoke/20260607T110958Z-progress-plus-tts-4b1019f4.json
  • .artifacts/telegram-smoke/20260607T111043Z-progress-long-task-25862d4a.json
• This is why the PR is draft: the additive TTS behavior is proven, but combined progress-plus-TTS live proof still needs a better progress trigger or manual GUI/transcript proof.

AI Assistance

• AI-assisted
• Testing degree: targeted + isolated live-tested