Bug type

Regression (worked before, now fails)

Summary

All proactive outbound sends via the message tool or TTS delivery fail with:

Error: No active WhatsApp Web listener (account: default).
This includes text messages sent via the tool, media, and audio from TTS. Auto-replies (text responses from the agent) work perfectly. Reactions also work.

Steps to reproduce

1. Set up WhatsApp channel (linked, connected, receiving messages works)
2. From the agent, call the message tool: action=send, channel=whatsapp, target=, message="test"
3. Error: No active WhatsApp Web listener
4. Auto-text replies from the agent work fine

Expected behavior

Agent should be able to send audio and video should through Whatsapp

Actual behavior

⚠️ ✉️ Message: /tmp/voice_reply.ogg failed

OpenClaw version

2026.3.12

Operating system

Linux (x64), Node v25.6.1

Install method

npm

Model

anthropic/claude-sonnet-4-6

Provider / routing chain

openclaw -> Antrhopic

Config file / key location

No response

Additional provider/model setup details

setActiveWebListener and requireActiveWebListener both import from dist/config-CmS8VEM4.js (same file path), but appear to operate on separate module instances at runtime — meaning setActiveWebListener writes to one listeners Map and requireActiveWebListener reads from a different one.

The auto-reply path (web-auto-reply → msg.reply() / msg.sendMedia()) uses a closure-captured socket reference and bypasses requireActiveWebListener entirely — which is why it works.

Patching dist/config-CmS8VEM4.js with console.error debug logs showed no output, confirming the gateway runtime is not loading the filesystem copy of that file (despite the package not declaring "sea").

Logs, screenshots, and evidence

Impact and severity

Affected: Whatsapp
Severity: High

Additional information

This worked previously. Not sure when this was broken, but it prevents Openclaw from sending images and audio to Whatsapp