Summary

Internal reasoning traces are intermittently delivered as user-visible messages in Telegram, e.g.:

• Reasoning: Planning change summary for approval
• Reasoning: Clarifying reminder type and timing
• Reasoning: Planning apology and bug explanation

These appear as separate chat messages before/around normal assistant responses.

Steps to reproduce

1. Update OpenClaw to 2026.2.22-2 (or start from a fresh session on that version).
2. Use Telegram direct chat with the assistant.
3. Send normal prompts that trigger multi-step handling (examples):

• ask for a reminder setup
• ask for a status check + follow-up action

4. Observe inbound Telegram messages around assistant replies.
5. Intermittently, a separate message appears starting with Reasoning: ... (internal planning text), e.g.:

• Reasoning: Clarifying reminder type and timing
• Reasoning: Planning apology and bug explanation

6. Confirm this text is user-visible in Telegram and not just internal logs.

Expected behavior

Reasoning/thought traces should never be delivered to end-user chat surfaces. Only final assistant content should be sent.

Actual behavior

Reasoning-prefixed internal text leaks into Telegram conversation as plain user-visible messages.

OpenClaw version

2026.2.22-2

Operating system

macOS 26.3 (arm64)

Install method

Node: 25.6.1

Logs, screenshots, and evidence

Impact and severity

Frequency
Multiple occurrences in one session, reproducible across several turns after update.

Impact

• Confusing UX for end users
• Exposes internal planning text
• Reduces trust in message quality and channel hygiene

Additional information