Empowering LLMs with Parameterized Skills for Adversarial Long-Horizon Planning