openclaw - 💡(How to fix) Fix [Bug]: Telegram group chat: streaming mode silently drops replies for models without adaptive thinking

Bug type

Behavior bug (incorrect output/state without crash)

Beta release blocker

No

Summary

When channels.telegram.streaming.mode is set to anything other than off, models that do not support adaptive thinking silently fail to deliver replies in Telegram group chats. The agent processes the message and generates a response internally, but nothing is ever posted to the group. No error is shown to the user.

Setting streaming.mode: "off" immediately fixes delivery for all affected models.

Steps to reproduce

1. Configure OpenClaw with Telegram group chat
2. Set streaming mode to anything other than off (partial, block, progress)
3. Set the active model to one without adaptive thinking (e.g. claude-haiku-4-5)
4. Send a message to the bot in the group
5. Observe: bot shows typing indicator, processes the message, but no reply is posted to the group

Expected behavior

Reply is delivered to the Telegram group regardless of streaming mode or thinking mode. In testing, claude-sonnet-4-6 with Think: adaptive delivers replies correctly under the same config.

Actual behavior

Reply is silently dropped. Internal session transcript shows the response was generated (systemSent: true), but no outbound sendMessage fires to Telegram. This matches the mechanism described in #66459.

OpenClaw version

2026.5.7

Operating system

Ubuntu 26.04

Install method

npm global

Model

anthropic/claude-haiku-4-5 (Think: medium) openai/gpt-4o xai/grok-3

Provider / routing chain

openclaw -> anthropic/claude-haiku-4-5 openclaw -> openai/gpt-4o openclaw -> xai/grok-3

Additional provider/model setup details

No response

Logs, screenshots, and evidence

Impact and severity

No response

Additional information

No response