Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

Summary

Internal/runtime formatting artefacts are occasionally being delivered as standalone Telegram messages in a direct chat. These messages are visible to the user but are not meaningful assistant output.

Examples observed

The following standalone messages appeared in Telegram:

• <channel|>
• set-thought <channel|>
• ───

Steps to reproduce

Not sure how to reproduce this exactly. This happened multiple times in the same direct Telegram conversation, so it does not look like a one-off rendering issue.

Expected behavior

Expected behaviour

Internal channel markers, debug placeholders, structural separators, or empty formatting artefacts should never be sent to user-facing messaging surfaces.

Actual behavior

These artefacts surfaced as normal Telegram messages.

OpenClaw version

2026.5.26 (10ad3aa)

Operating system

macOS

Install method

npm global

Model

gpt 5.5

Provider / routing chain

codex

Additional provider/model setup details

No response

Logs, screenshots, and evidence

Impact and severity

No response

Additional information

No response