Text this: Reinforcement Learning for Fail-Operational Systems with Disentangled Dual-Skill Variables