Description:
The OpenClaw webchat interface is incorrectly rendering the agent's internal monologue instead of the user-facing block. This makes it appear as if the agent is not responding to the user, and only showing its internal thoughts, making the interface nearly unusable.
Steps to Reproduce:
Use the standard OpenClaw webchat interface.
Have a conversation with an agent.
Observe that the agent's response in the chat window is only the content of its block. The intended response is not displayed at all.
Inform the agent that you can only see its thoughts (e.g., "I can only see your thoughts").
Observe that in the subsequent turn, the agent's response is often displayed correctly, in which it explains the vs issue.
Expected Behavior:
The webchat interface should always parse the agent's raw output and only display the content within the block to the end-user. The block should always be hidden.
Actual Behavior:
The user only sees the content of the block. For example, they will see this:
I am planning my next action here. I will use the 'read' tool to get the file content.
...instead of the intended final response, which is completely hidden.
When the user points out this problem, the agent's next response (which typically explains the bug) is often displayed correctly, showing the block as expected. This inconsistent, stateful behavior is a key part of the issue.
Environment Details:
OpenClaw Version: 2026.1.30
Agent Model: google-gemini-cli/gemini-2.5-pro
Host OS: Ubuntu
Interface: Webchat
Additional Context:
This issue makes the webchat interface nearly unusable, as the user is only seeing the agent's internal reasoning, not its actual replies. The fact that the rendering behavior changes after the user reports the problem suggests a potential issue with how the webchat channel handles state or parses the agent's output on different turns.
Description:
The OpenClaw webchat interface is incorrectly rendering the agent's internal monologue instead of the user-facing block. This makes it appear as if the agent is not responding to the user, and only showing its internal thoughts, making the interface nearly unusable.
Steps to Reproduce:
Use the standard OpenClaw webchat interface.
Have a conversation with an agent.
Observe that the agent's response in the chat window is only the content of its block. The intended response is not displayed at all.
Inform the agent that you can only see its thoughts (e.g., "I can only see your thoughts").
Observe that in the subsequent turn, the agent's response is often displayed correctly, in which it explains the vs issue.
Expected Behavior:
The webchat interface should always parse the agent's raw output and only display the content within the block to the end-user. The block should always be hidden.
Actual Behavior:
The user only sees the content of the block. For example, they will see this:
I am planning my next action here. I will use the 'read' tool to get the file content.
...instead of the intended final response, which is completely hidden.
When the user points out this problem, the agent's next response (which typically explains the bug) is often displayed correctly, showing the block as expected. This inconsistent, stateful behavior is a key part of the issue.
Environment Details:
OpenClaw Version: 2026.1.30
Agent Model: google-gemini-cli/gemini-2.5-pro
Host OS: Ubuntu
Interface: Webchat
Additional Context:
This issue makes the webchat interface nearly unusable, as the user is only seeing the agent's internal reasoning, not its actual replies. The fact that the rendering behavior changes after the user reports the problem suggests a potential issue with how the webchat channel handles state or parses the agent's output on different turns.