Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

After upgrading from 2026.4.15 to 2026.4.22 I started getting errors in telegram (every 2nd-3rd message randomly)

⚠️ Agent couldn't generate a response. Please try again.

Steps to reproduce

1. start openclaw 2026.4.22+
2. chat with the bot in telegram
3. bot replies show "⚠️ Agent couldn't generate a response. Please try again."

Expected behavior

Proper chat, messages get delivered

Actual behavior

Logs show this:
incomplete turn detected: runId=b73b8dbf-d6d4-410e-a23b-9bc06173005f sessionId=2170070b-5108-4f21-a21a-9f0779d09267 stopReason=stop payloads=0 — surfacing error to user

I'm using llama.cpp as a backend, configured as openai-completions in the config.

Judging from llama.cpp logs, the message was generated in full:

srv  log_server_r: done request: POST /chat/completions 127.0.0.1 200
slot print_timing: id  0 | task 3932 |
prompt eval time =     632.61 ms /   950 tokens (    0.67 ms per token,  1501.71 tokens per second)
       eval time =    1786.86 ms /   103 tokens (   17.35 ms per token,    57.64 tokens per second)
      total time =    2419.47 ms /  1053 tokens
slot      release: id  0 | task 3932 | stop processing: n_tokens = 30311, truncated = 0
srv  update_slots: all slots are idle

However, openclaw treats it as an incomplete message - thus fully breaking it for me.
I didn't see this behavior on 2026.4.15 - it was working flawlessly with this backend.

OpenClaw version

2026.4.22

Operating system

ubuntu

Install method

docker

Model

Qwen3.6-27B, Qwen3.6-35B-A3B, Gemma4-31B

Provider / routing chain

openclaw -> llama.cpp

Additional provider/model setup details

No response

Logs, screenshots, and evidence

Impact and severity

No response

Additional information

No response