This does not normalize for tokens used if their skill description was as large as the docs index and contained all the reasons the LLM might want to use the sk

This does not normalize for tokens used if their skill description was as large as the docs index and contained all the reasons the LLM might want to use the skill, it likely performs much better than just one sentence as well.