Is the skill.json generated by LLM or obtained by other means? Will you open source the construction method of skill.json in the future?