[Feature Request] Add decorator for browser toolkit to ask agent reasoning while action

[nitpicker55555]

Required prerequisites

• I have searched the Issue Tracker and Discussions that this hasn't already been reported. (+1 or comment there if it has.)
• Consider asking first in a Discussion.

Motivation

I found that although send_message_to_user in the current workforce provides the ability to send messages during an action, it does not store those messages in the context. Therefore, we need an additional decorator to enable reasoning during actions and insert the generated content into the context.
To avoid affecting other tools, we can first create such a decorator specifically for the browser_toolkit and test whether it improves performance.

Solution

No response

Alternatives

No response

Additional context

No response

[Wendong-Fan]

thanks @nitpicker55555 for opening this issue, I think send_message_to_user would be recorded as tool call history in the context, for the feature we want to implement that sounds like interleaved thinking, refer to : https://platform.claude.com/docs/en/build-with-claude/extended-thinking#interleaved-thinking, does this still necessary for us to implement this feature in agent level as now many models are supporting this feature (e.g. gemini 3 pro, claude, minimax m2 etc)

[fengju0213]

thanks @nitpicker55555 for opening this issue, I think send_message_to_user would be recorded as tool call history in the context, for the feature we want to implement that sounds like interleaved thinking, refer to : https://platform.claude.com/docs/en/build-with-claude/extended-thinking#interleaved-thinking, does this still necessary for us to implement this feature in agent level as now many models are supporting this feature (e.g. gemini 3 pro, claude, minimax m2 etc)

Perhaps we could provide a decorator — similar to message integration — that adds a parameter called reasoning to tools that need it, and see if it leads to performance improvements.

[Ishneet0710]

The current implementation follows @fengju0213's suggestion - it provides a decorator that adds a reasoning parameter to browser tools. The reasoning gets stored in the agent's context (not just logged), which addresses the original issue about send_message_to_user not storing in context.

[YunfeiZHAO]

PR has been added for the current browser tool.
@Ishneet0710, I have reviewed your PR, but the update is for the deprecated browser toolkits. We should have a better discussion during the meeting. Sorry for the misunderstanding.
Also, direct modifications to the agent's memory are not very safe.
Please review #3552 and let's discuss tomorrow!
Thank you for the PR!