Supervisor Architecture - How do I directly call a tool without any processing by the agent

[Curiosity007]

I have this hierarchy maintained in my codebase - supervisor, agent a, tool a.

I am using create_supervisor method to create the supervisor with required agents, tools and custom handoff tools.

Now, the problem statement -

Supervisor gets entire message state and knows when to call agent a. Now, this agent a, should call tool a.

The problem I am facing, agent a is doing one layer of prompt processing, and then passing the paraphrased output to tool a.

What I want is, that tool a , should get the data directly, unchanged. Helps with latency, token etc.

How do I accomplish this