Description

When thinking=low (or any thinking mode) is enabled, the model's reasoning/thinking blocks are being sent to WhatsApp as part of the visible message. They should be stripped before delivery.

Expected behavior

Reasoning blocks should be hidden from the user on all channels, as documented: "Reasoning: on (hidden unless on/stream)".

Actual behavior

The full reasoning text is included in the WhatsApp message, visible to the end user. Example:

Reasoning:
_The reasoning is leaking again..._

This appears appended to the actual reply.

Environment

• OpenClaw version: 2026.2.22-2
• Channel: WhatsApp
• Model: anthropic/claude-opus-4-6
• Thinking: low

Steps to reproduce

1. Enable thinking (thinking=low)
2. Send a message via WhatsApp
3. The reasoning block appears in the delivered WhatsApp message