One of the major miss right now seems to be in tool calling specs, you specify function names, description, inputs but not outputs. I believe with reasoning models planning things, it would be important to understand output format, descriptions as well?