I don't think the idea of skills is quite snake oil. It seems you can change what LLM outputs next by what's called few-shot prompting or in-context learning: https://www.promptingguide.ai/techniques/fewshot