Summary

In webchat, starting a new session displays internal bootstrap/system text and also shows literal reply control tags (e.g., [[reply_to_current]]) in assistant output.

Steps to reproduce

1. Start a fresh conversation with /new (or /reset).
2. Let the assistant send its first greeting.
3. Observe the transcript in webchat.
4. Internal system instruction text appears as a visible message.
5. Assistant message includes literal [[reply_to_current]] prefix.

Expected behavior

Internal/system/bootstrap instructions are not shown in user-visible chat. Reply control tags are stripped before storage/render/send, so users only see normal assistant text.

Actual behavior

A message containing internal /new instruction text is visible in the transcript. The assistant greeting starts with [[reply_to_current]] rendered literally. This is confusing and exposes internal control/meta content.

OpenClaw version

Observed on build current as of 2026-02-24 (exact app version not shown in UI at report time).

Operating system

Linux (OpenClaw host), webchat surface in Chrome

Install method

No response

Logs, screenshots, and evidence

Impact and severity

No response

Additional information

No response