We propose InDiGO, a knowledge-aware, diversity-optimized framework that aligns clinical signals with decision cues to iteratively refine series-text prompts, enabling effective and generalizable medical time-series decoding with LLMs.