BUG: The @browser-use agent returns 'null' in almost all results

[Fahell]

No matter which model is selected, the agent returns "null" in virtually all calls. It succeeds in tool calls and then suddenly returns null. It seems to me that it is passing incorrect parameters to some tool call or something similar, which returns this "null" value and prematurely ends the agent's interaction.

[Fahell]

Another example that makes me think that perhaps the instructions or the LLM model used for the agent (That is, if the model used for that agent is not the main model) aren't sufficient for the task. You can see the agent passing invalid arguments during the 'key' call, and right below that, another 'null'.