This study proposes the Theme-Content Consistency Protocol (TCCP) as a novel defense mechanism against prompt injection attacks on Large Language Models (LLMs). TCCP evaluates the consistency between the declared theme and actual content of user inputs, invalidating requests with structural inconsistencies. Unlike conventional keyword-based or pattern-matching approaches, TCCP detects structural contradictions, demonstrating high resilience against novel attack patterns. Through a 12-month observational study by an independent researcher, TCCP was found to not only defend against prompt injection but also improve the overall logical consistency and response quality of LLMs by 25-35%. Implementation requires only system-level principle definition without complex additional infrastructure. This paper reports the theoretical foundation of TCCP, evaluation results, and unexpected secondary effects.Co-written by Viorazu. and Claude ( Sonnet 4.5, Anthropic)共著:Viorazu. & Claude( Sonnet 4.5、Anthropic)