openclaw - 💡(How to fix) Fix Telegram /reasoning stream not showing character-by-character (regression in v2026.5.7)

Bug Description

Environment:

• OpenClaw version: v2026.5.7
• Platform: macOS (Darwin)
• Channel: Telegram
• Model: MiniMax-M2.7

Problem: On v2026.4.x, executing /reasoning stream in Telegram caused the reasoning content to appear character by character in a draft bubble, then disappear after completion. This behavior was confirmed working on v2026.4.15 (April 4, 2026).

After upgrading to v2026.5.7, the same command results in reasoning appearing as a full block (all content at once) rather than streaming character-by-character. The reasoning still disappears after completion, but it no longer streams progressively.

Additional observations:

• This happens regardless of channels.telegram.streaming.mode setting (partial, off, etc.)
• Tested with mode=off + /reasoning stream: reasoning appears as one block then disappears
• Tested with mode=partial + block.enabled=true: same behavior (no character-by-character output)
• Message loss (reported in #80520) is a separate but possibly related issue

Expected behavior (from v2026.4.x):

1. Reasoning content streams character-by-character into Telegram draft bubble
2. Draft bubble updates incrementally as reasoning generates
3. Draft disappears after completion, final answer sent

Actual behavior (v2026.5.7):

1. No draft bubble appears during generation
2. Reasoning appears as a complete block when model finishes
3. Block disappears, final answer sent

Related issues:

• #80520 (Telegram message loss - streaming regression)

Commands tested:

• /reasoning stream - same broken behavior
• /reasoning on - shows reasoning but not streaming

Is this regression known? Any workaround?