Problem

When an agent intends to suppress a message (e.g., after sending via the message tool, or when a cron result doesn't need user delivery), it should reply with NO_REPLY as the entire message.

In practice, LLMs frequently append NO_REPLY to the end of a real message — e.g.:

File's there. Same false failure as before — cron reports error but the write actually succeeded. Not urgent.

NO_REPLY

This gets sent to the user as-is, which looks broken.

Suggestion

Strip or suppress the message if NO_REPLY appears anywhere in the response (not just when it's the entire message). Or at minimum, strip the literal NO_REPLY token from the end of messages before delivery.

Alternatively, a config option like messages.stripNoReply: true that auto-cleans this.

Context

Happened twice in one day (March 1, 2026) on Telegram. The agent knows the rule but the LLM still appends it to real content occasionally. This feels like it should be handled at the system level rather than relying on prompt compliance.

Environment

• OpenClaw v2026.2.26
• Channel: Telegram
• Agent: main (Claude Opus 4.6)