Bug report

Internal assistant/planning text is being delivered to end users in chat, even when verbose mode is disabled.

What happened

A user received internal-looking messages such as:

• Need commit push.
• Need transcribe and reply likely content.

These lines look like internal planning/checklist text and should never be user-visible in normal chat output.

Expected behavior

• With /verbose off, only user-facing assistant responses should be sent.
• Internal planning/tool orchestration text should remain private.

Actual behavior

• Internal text appeared as normal bot messages in Telegram chat.

Environment

• OpenClaw: 2026.3.13
• Channel: Telegram
• Runtime observed from OpenClaw agent flow in direct chat

Repro hints

1. Keep /verbose off and /reasoning off.
2. Perform multi-step/tool-heavy actions (edits/commits/transcription, etc.).
3. Occasionally internal checklist/planning text is sent to user.

Impact

• Privacy/UX issue: internal operational text leaks into user-visible conversation.
• Confusing and reduces trust.

If useful, I can provide additional timestamps / sanitized excerpts.