Turn LLM instructions (skills, plugins, prompts) into standalone, deterministic scripts — more reliable than agent calls, with no token cost and faster execution.