openclaw - 💡(How to fix) Fix Bug: Fallback models echo BOOT.md instructions instead of executing them [1 participants]

Fix Plan

To address the issue, we will implement the second proposed solution: Suprimir la respuesta (Suppress the response). We will modify the OpenClaw code to detect if the model's response is a literal echo of the startup instructions and suppress it.

Step-by-Step Solution

• Modify the response processing function to compare the model's response with the startup instructions.
• If the responses match, suppress the model's response to prevent it from being sent to the user.

Example Code

def process_response(model_response, startup_instructions):
    # Remove leading/trailing whitespace and compare responses
    if model_response.strip() == startup_instructions.strip():
        # Suppress the response if it matches the startup instructions
        return None
    else:
        # Return the response if it does not match
        return model_response

# Example usage:
startup_instructions = "BOOT.md startup instructions"
model_response = "BOOT.md startup instructions"

processed_response = process_response(model_response, startup_instructions)
if processed_response is None:
    print("Response suppressed")
else:
    print("Response:", processed_response)

Verification

To verify the fix, test the modified code with different model responses, including literal echoes of the startup instructions. The response should be suppressed when it matches the startup instructions, and the user should not receive the internal system prompts.

Extra Tips

• Ensure that the comparison is case-sensitive and whitespace-sensitive to avoid false positives.
• Consider logging the suppressed responses for debugging and monitoring purposes.