This survey consolidates theoretical and empirical perspectives on in-context learning, covering mechanistic explanations, interpretability tools, and evaluation methodologies. We highlight open challenges and suggest future directions for understanding context-driven model behavior.