This keeps each node's view honest ("what you see is what you can do") and stops agents from wasting round-trips planning calls that'll just be rejected. Besides, it also improves model tool-selection since forbidden tools are pure noise in the decision space
To be discussed: Optionally allow administrators to reveal tools a node isn't authorized to call (can be useful for debugging purposes)
This keeps each node's view honest ("what you see is what you can do") and stops agents from wasting round-trips planning calls that'll just be rejected. Besides, it also improves model tool-selection since forbidden tools are pure noise in the decision space
To be discussed: Optionally allow administrators to reveal tools a node isn't authorized to call (can be useful for debugging purposes)