openclaw - 💡(How to fix) Fix Bug: Telegram DM voice can bypass transcription in followup queue

Summary

Telegram DM voice notes can still reach the agent as raw <media:audio> when they arrive while an agent run is already active and the deployment routes Telegram busy-run messages through followup.

This looks similar to #44682, but that issue is locked and was closed as not reproducible on current main. I am filing this as a fresh downstream report because the observed affected deployment is 2026.5.20 (e510042), and I have not yet retested against v2026.5.26 / current main.

Environment

• OpenClaw: 2026.5.20 (e510042)
• Channel: Telegram DM
• Queue config:

messages:
  queue:
    byChannel:
      telegram: followup

• tools.media.audio.enabled: true
• Audio transcription is configured with a local CLI first and an OpenAI fallback.

What I checked

• The local audio transcription CLI works when invoked directly on the same kind of Telegram .ogg voice file.
• Telegram DM voice preflight works for the normal immediate path.
• The remaining failure shape appears to be specifically the busy-run / queued-followup path: the followup turn can still see a raw audio placeholder instead of an [Audio transcript ...] body.

Expected behavior

Queued Telegram followup turns should receive the same media-understood body as the immediate inbound path.

For an audio-only Telegram DM voice note, the prompt should contain the generated transcript and should not expose raw <media:audio> to the model.

Actual behavior

When the voice note arrives while another agent run is active and Telegram uses followup, the later followup turn can receive raw <media:audio> without the transcript.

Suggested regression coverage

1. Configure Telegram DM voice transcription and messages.queue.byChannel.telegram = "followup".
2. Start an agent run and keep it active.
3. Send a Telegram DM voice note while the run is active.
4. Drain the queued followup turn.
5. Assert the prompt contains an audio transcript and does not contain raw <media:audio>.

Related

• #44682: exact earlier failure shape for followup-queued audio media understanding
• #61008: fixed Telegram DM voice-note preflight for the immediate path

Workaround

Switch Telegram busy-run routing from followup to steer until the queued-media path is verified.