Summary

Fixes #29023. WhatsApp reply-to-bot detection (_message_is_reply_to_bot) consistently returned False on multi-device accounts because of a normalized-ID shape mismatch:

• Baileys delivers quotedParticipant without the device suffix: "5511999999999@s.whatsapp.net"
• sock.user.{id,lid} populates botIds with the device suffix (:N → normalized to @N): "5511999999999@10@s.whatsapp.net"

The set-membership check in gateway/platforms/whatsapp.py therefore never matched, breaking the reply-to-bot admission shortcut in groups configured with require_mention: true. This is especially painful for voice notes, where WhatsApp doesn't support text @mentions at all — there's literally no other way to wake the bot.

Fix

In _bot_ids_from_message, when a normalized bot id has three @-separated parts (i.e. carries a device suffix), add a suffix-stripped alias alongside the original. Legacy two-part ids are unchanged, so single-device accounts keep their existing behavior.

parts = normalized.split("@")
if len(parts) > 2:
    bot_ids.add(f"{parts[0]}@{parts[-1]}")

This is exactly the patch the issue author proposed in the Fix section.

Tests

Added two regression tests in tests/gateway/test_whatsapp_group_gating.py:

• test_reply_to_bot_matches_device_suffixed_bot_ids — multi-device botIds (:10 suffix) + non-suffixed quotedParticipant must match and admit the message under require_mention=true.
• test_reply_to_bot_legacy_non_multidevice_unchanged — legacy two-part ids continue to match as before.

$ python -m pytest tests/gateway/test_whatsapp_group_gating.py -q
28 passed in 2.52s

Risk

Tiny, surgical, single-function change. The added alias is only emitted when the id has a third @-separated component, so it cannot affect legacy callers. No other code paths consume _bot_ids_from_message in a way that would break on the extra alias (the other callers use it for split('@', 1)[0] bare-id extraction and in-checks against mentionedIds).