Description

When the agent processes certain messages (particularly in group chats or after cron jobs), the internal reasoning text (marked with </think> tags) is being sent to the user as a regular message instead of being filtered out.

Steps to Reproduce

1. Have a cron job complete and send results to a WhatsApp group
2. Or send a message that triggers internal reasoning
3. The user receives two messages: one with the actual reply, and one with the raw reasoning text

Expected Behavior

Internal reasoning (the content between </think> and <final> tags in the model's output) should never be sent to users.

Actual Behavior

Users receive messages like:

Reasoning:
Sessions file is empty...

抱歉，我看不到...

The internal reasoning is being exposed as a message.

Environment

• OpenClaw version: 2026.2.22-2
• Channel: WhatsApp
• OS: macOS

Additional Context

This has been observed in WhatsApp group chats. The issue may be related to how the model output is processed after cron jobs or certain triggers.