### Thinking Token Sanitization- **Newer OpenClaw versions strip thinking
✅Solved
fix: strip reasoning tags from messaging tool text to prevent <think> leakage (#11053)
Hello,
I like this, it fixed thinking leaking into chat at times,
but after this Kimi K2.5 Thinking was actually silent, I debugged it together with non Thinking:
it tended to often send a whole reply in "thinking" and that got stripped out:
I had it add to AGENTS.md
"## OpenClaw Runtime Behavior
- This prevents non-output tokens from leaking into user-facing messages
- If the model only generates thinking without actual output content, the message will be empty/silent
- When using thinking models, ensure actual response content is generated, not just internal reasoning
"
and then it worked and talked normally
Hello,
I like this, it fixed thinking leaking into chat at times,
but after this Kimi K2.5 Thinking was actually silent, I debugged it together with non Thinking:
it tended to often send a whole reply in "thinking" and that got stripped out:
I had it add to AGENTS.md
"## OpenClaw Runtime Behavior
Thinking Token Sanitization
- Newer OpenClaw versions strip thinking/reasoning tokens from chat output- This prevents non-output tokens from leaking into user-facing messages
- If the model only generates thinking without actual output content, the message will be empty/silent
- When using thinking models, ensure actual response content is generated, not just internal reasoning
"
and then it worked and talked normally
Solution
To help others find answers, you can mark your question as solved via
Right click solution message -> Apps -> ✅ Mark Solution
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
83,815Members
Resources
Was this page helpful?
