Voice Memo

Send native iMessage voice bubbles with ElevenLabs TTS via BlueBubbles. Use when: user asks to send a voice message, wants something spoken aloud, storytelli...

Security Scan

The skill's code and runtime instructions match its stated purpose (sending iMessage voice bubbles via ElevenLabs + BlueBubbles); the main concern is an authoring/metadata inconsistency and the usual need to trust local env/config values it reads.

Purpose & Capability

The script implements exactly what the skill describes: it calls ElevenLabs TTS, converts audio with afconvert to Opus CAF, and posts to a BlueBubbles Private API to produce native iMessage voice bubbles. Required binaries (curl, afconvert) and the two service credentials are appropriate for this functionality.

Instruction Scope

SKILL.md and the script confine actions to generating audio, converting it, and sending it to the BlueBubbles endpoint. The script sources ~/.openclaw/.env for credentials and does not attempt to read unrelated system files or exfiltrate arbitrary data. It documents how transcriptions integrate into conversation context and notes that transcriptions are not auto-persisted.

Install Mechanism

No install spec (instruction-only) and a small included shell script. No downloads, archive extraction, or third-party package installs — low install-time risk.

Credentials

The script requires ELEVENLABS_API_KEY and BLUEBUBBLES_PASSWORD (proportionate to the task). However, registry metadata at the top of the submission lists no required env vars while SKILL.md and the script do — an inconsistency. Also note the script blindly sources ~/.openclaw/.env, which may contain additional secrets; ensure that file contains only intended credentials and is protected.

Persistence & Privilege

always:false and no system-wide configuration changes. The skill can be invoked autonomously (platform default), which is expected for skills; it does not request persistent elevated privileges or modify other skills' configs.

Assessment

This skill appears to do what it claims: generate ElevenLabs TTS, convert to Opus CAF, and post to a BlueBubbles Private API to create native iMessage voice bubbles. Before installing: (1) verify the skill's source (homepage is a GitHub repo) and inspect ~/.openclaw/.env to ensure it only contains the expected ELEVENLABS_API_KEY and BLUEBUBBLES_PASSWORD; (2) ensure BLUEBUBBLES_URL points to a trusted local BlueBubbles instance (default is http://127.0.0.1:1234); (3) treat the ElevenLabs API key as billable/privileged — restrict its scope and rotate if shared; (4) be aware that if you set BLUEBUBBLES_URL to a remote endpoint, audio and metadata will be sent there (so only use trusted endpoints); and (5) note the metadata/registry inconsistency (registry omitted required env vars) — this is likely a packaging oversight but worth confirming with the author before granting credentials.

Send native iMessage voice bubbles (not file attachments) using ElevenLabs TTS and BlueBubbles.

Quick Start

Run the script with text and recipient:

scripts/send-voice-memo.sh "Your message here" [phone]

This will:

1. Generate TTS audio via ElevenLabs (Rachel voice by default)
2. Convert to Opus CAF @ 24kHz (iMessage native format)
3. Send as native voice bubble via BlueBubbles

Requirements

• BlueBubbles running locally with Private API enabled
• ElevenLabs API key (for TTS)
• macOS (for afconvert audio conversion)
• Environment variables in ~/.openclaw/.env:

  ELEVENLABS_API_KEY=your-key-here
  BLUEBUBBLES_PASSWORD=your-password-here
  # Optional overrides:
  ELEVENLABS_VOICE_ID=21m00Tcm4TlvDq8ikWAM  # Rachel (default)
  ELEVENLABS_MODEL_ID=eleven_turbo_v2_5      # Turbo v2.5 (default)
  

The Working Formula

Critical parameters discovered 2026-02-19:

Parameter	Value	Why
chatGuid	any;-;+PHONE	NOT iMessage;-; (causes timeouts)
method	private-api	Required for native bubble
isAudioMessage	true	Required
Audio format	Opus @ 24kHz in CAF	iMessage native format
Pre-convert	Yes	Don't let BlueBubbles convert (wrong codec)

Voice Options

Default voice: Rachel (ElevenLabs)

• Voice ID: 21m00Tcm4TlvDq8ikWAM
• Model: eleven_turbo_v2_5 (fast, natural)
• Cost: ~$0.04 per 30s message

Expressive tags:

• [laughs] — natural laughter
• [sighs] — expressive sigh
• [excited] — energetic delivery

Example: "[excited] Oh my god, it worked!"

For full voice list and IDs, see VOICES.md.

Bidirectional Voice Memos

Sending (Amz → Amy): Use this skill. Native voice bubbles appear with waveform UI.

Receiving (Amy → Amz): BlueBubbles auto-converts incoming voice memos to MP3. OpenClaw transcribes via Whisper. Transcribed text flows into conversation context automatically.

Memory note: Incoming voice memo transcriptions flow into conversation context like any text message. They are NOT automatically persisted to memory or files — the agent must explicitly choose to store them, same as any conversation content. If you want to prevent transcriptions from being retained, instruct the agent not to record voice memo content in memory.

Troubleshooting

Voice bubble arrives as file attachment:

• Check method=private-api is set
• Verify chatGuid uses any;-; prefix (not iMessage;-;)
• Check response has "isAudioMessage": true

API times out:

• Use any;-;+PHONE format for chatGuid
• Verify BlueBubbles Private API is enabled
• Restart BlueBubbles if consistently slow

Audio is 0 seconds / unplayable:

• Ensure pre-conversion to Opus @ 24kHz
• Don't let BlueBubbles convert (uses wrong codec)
• Verify with: afinfo output.caf (should show opus @ 24000 Hz)