openclaw - 💡(How to fix) Fix lossless-claw + thinking + tool-call triggers 400 invalid_request_error, causing fallback [1 participants]

400 invalid_request_error: "thinking is enabled but reasoning_content is missing in assistant tool call message at index 3"

TL;DR

The issue can be resolved by ensuring that the reasoning_content field is preserved in assistant tool-call messages when using the lossless-claw context engine with thinking enabled.

Guidance

• Verify that the reasoning_content field is being sent correctly in the tool-call messages by checking the request payload before it is processed by lossless-claw.
• Check the lossless-claw configuration to ensure that it is not intentionally dropping or modifying the reasoning_content field.
• Consider modifying the lossless-claw implementation to preserve the reasoning_content field, or use a different context engine that handles this field correctly.
• Test the task execution with a different context engine, such as memory-core, to confirm that the issue is specific to lossless-claw.

Example

No code snippet is provided as the issue is related to the configuration and implementation of the lossless-claw context engine.

Notes

The issue seems to be specific to the lossless-claw context engine and the way it handles assistant tool-call messages with thinking enabled. The memory-core context engine does not exhibit this behavior, suggesting that the issue lies in the lossless-claw implementation.

Recommendation

Apply a workaround by modifying the lossless-claw implementation to preserve the reasoning_content field, or use a different context engine that handles this field correctly, as the current implementation is causing the fallback to the secondary model.