Text this: Safety Context Injection: Inference-Time Safety Alignment via Static Filtering and Agentic Analysis