Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

Upgrading from 2026.4.23 to 2026.4.29 on Windows causes three related failures:

1. WhatsApp WebSocket drops with 408 Request Time-out every ~60 seconds, killing connections before the agent can deliver replies
2. openclaw channels login --channel whatsapp hangs indefinitely after loading plugins — never displays QR code, cannot be interrupted with Ctrl+C
3. Agent replies never delivered — inbound messages are received and ack'd, but the agent response is lost because the WhatsApp connection drops before delivery completes

Rolled back to 2026.4.23 where all three issues are absent.

Steps to reproduce

1. Running stable OpenClaw 2026.4.23 on Windows with WhatsApp channel working normally
2. Gateway runs as Windows Scheduled Task
3. WhatsApp linked to dedicated number (+47...) with dmPolicy: allowlist

Expected behavior

• WhatsApp connection remains stable for extended periods (as it does on 2026.4.23)
• channels login displays QR code for pairing within 30 seconds
• Agent replies are delivered reliably after receiving inbound messages
• Bonjour failures do not impact WhatsApp or agent functionality

Actual behavior

WhatsApp 408 loop (from gateway log)

[gateway/channels/whatsapp] [default] starting provider (+47...)
[gateway/channels/whatsapp] Listening for personal WhatsApp inbound messages.
# ~60 seconds later:
[web-reconnect] connection closed {status: 408, loggedOut: false, reconnectAttempts: 1, error: "status=408 Request Time-out Connection was lost"}
[gateway/channels/whatsapp] WhatsApp Web connection closed (status 408). Retry 1/12 in 2.33s…
# Reconnects, then drops again within 60 seconds


### OpenClaw version

2026.4.29 (a448042)

### Operating system

Windows 11 Home (10.0.26200, x64)

### Install method

`npm install -g [email protected]`

### Model

openai/gpt-5

### Provider / routing chain

openclaw -> openai (API key, direct)

### Additional provider/model setup details

Main agent: openai/gpt-5 with openai/gpt-4o fallback.
Cron agent: openai/gpt-4o-mini with anthropic/claude-haiku-4-5 fallback.
Gateway: local loopback (127.0.0.1:18789), auth: token, Windows Scheduled Task.
WhatsApp: single account, dmPolicy: allowlist, dedicated number.
Config file unchanged between versions (6249 bytes verified before and after upgrade).
Note: The bugs are channel/gateway-level — not model-dependent.

### Logs, screenshots, and evidence

```shell
Key log lines from C:\tmp\openclaw\openclaw-2026-05-02.log (redacted):

**Gateway startup and WhatsApp connect:**
[gateway] http server listening (3 plugins: active-memory, memory-core, whatsapp; 21.2s)
[gateway/channels/whatsapp] [default] starting provider (+47...)
[gateway/channels/whatsapp] Listening for personal WhatsApp inbound messages.

**Message received but reply never delivered:**
[web-inbound] inbound message {from: "+47...", body: "Hei. Er du der?", timestamp: 1777716752000}
[web-auto-reply] sending ack reaction {emoji: "👍"}
[gateway/channels/whatsapp/inbound] Inbound message +47... -> +47... (direct, 78 chars)
# No "auto-reply sent" ever appears

**408 timeout ~60s after connect:**
[web-reconnect] connection closed {status: 408, loggedOut: false, reconnectAttempts: 1, error: "status=408 Request Time-out Connection was lost"}
[gateway/channels/whatsapp] WhatsApp Web connection closed (status 408). Retry 1/12 in 2.48s…

**`channels login` hang (reproduced 3 times):**
11:55:01 [plugins] loaded 117 plugin(s) (72 attempted) in 65265.2ms
# No further output. Hangs indefinitely. Ctrl+C ineffective.

**Bonjour failure:**
[plugins] bonjour: disabling advertiser after 3 failed restarts (service stuck in unannounced for 20026ms (gateway fqdn=Lenono_Reserve (OpenClaw)._openclaw-gw._tcp.local.))

**Config-triggered self-restart on first boot after upgrade:**
[gateway/reload] config change requires gateway restart (plugins.entries.active-memory.config.model)

**After rollback to 4.23 — immediately stable:**
openclaw channels status --probe → WhatsApp default: enabled, configured, linked, running, connected

Impact and severity

Affected: Windows users running WhatsApp channel on 2026.4.29 Severity: Blocks workflow — WhatsApp channel completely non-functional (messages received but never answered) Frequency: Always — every WhatsApp connection drops within ~60 seconds, reproduced across multiple gateway restarts and credential resets over 3 hours of testing Consequence: Bot appears unresponsive to all WhatsApp users. Agent replies are never delivered. channels login cannot re-pair after credential reset (hangs indefinitely without QR code). Only workaround is full rollback to 2026.4.23.

Additional information

Last known good version: 2026.4.23 (a979721) First known bad version: 2026.4.29 (a448042) Versions 2026.4.25 and 2026.4.27 also had WhatsApp issues on this machine (stuck session, event loop blocked at 91% CPU) — both rolled back to 2026.4.23.

Root cause issues already filed:

• #75804 — qrcode missing from plugin-runtime-deps (explains why channels login hangs silently on Windows — QR render fails without error output to terminal)
• #75610 — severe event loop block, API timeout, stuck session & Bonjour mDNS loop

Related issues with overlapping symptoms:

• #75717 — TUI hangs on 4.29 (WebSocket handshake timeout, 50-92% CPU)
• #75430 — TUI startup hangs at 100% CPU due to plugin reload loop (4.29)
• #72355 — Bonjour/mDNS hardcoded hostname causes crash loop

Windows-specific notes:

• channels login fails silently — no error output, no QR code, Ctrl+C ineffective (must close PowerShell window). Linux users see the qrcode error in terminal; Windows users see nothing.
• Hostname contains underscore (LENONO_RESERVE) which may contribute to Bonjour mDNS failures.
• Gateway runs as Windows Scheduled Task via gateway-start.cmd wrapper.