How Skills Affect Costs
Each skill/tool definition adds tokens to every request. Models with prompt caching
can cache these definitions — paying a cheap cache-read rate per request instead of full input price.
Models without caching pay full input price for tool schemas on every call.