Comment by storystarling

I tried this recently and found the token overhead makes it prohibitive for any non-trivial schema. Dumping the full introspection result into the context window gets expensive fast and seems to increase hallucination rates compared to just providing specific, narrow tool definitions.