Problem

When a user sends a voice message, OpenClaw transcribes it (via whisper or configured provider) and includes the transcript in the [Audio] User text: ... Transcript: prefix. However, there's no way to automatically echo the transcription back to the chat before the agent processes and responds.

This means the user has to trust the agent will echo the transcript — but this is a prompt-based behavior that agents forget inconsistently. The user wants to confirm what was heard before the agent acts on it.

Proposed Solution

A config option under tools.media.audio (or agents.defaults):

{
  "audio": {
    "echoTranscript": true,
    "echoFormat": "📝 *\"{transcript}\"*"
  }
}

When echoTranscript is true, the gateway automatically sends the transcription text back to the originating chat as a reply/quote before forwarding the message to the agent session. This is infrastructure-level — zero agent involvement, 100% reliable.

Why This Matters

• Voice messages are inherently ambiguous — the user needs to verify what was transcribed
• Agent prompt instructions ("always echo transcription first") are unreliable — agents forget
• This is a common workflow: voice → confirm transcript → agent processes
• Making it a gateway-level feature means it works for all agents, all channels, automatically

Alternatives Considered

• Prompt instructions in AGENTS.md — unreliable, agents forget
• Skills/hooks — OpenClaw doesn't have message preprocessing middleware
• Custom commands — doesn't help with organic voice messages