Large Language Model (LLM) agents are increasingly being deployed as conversational assistants capable of performing complex real-world tasks through tool integration. This enhanced ability to interact with external systems and process various data sources, while powerful, introduces significant security vulnerabilities. In particular, indirect prompt injection attacks pose a critical threat, where malicious instructions embedded within external data sources can manipulate agents to deviate from user intentions. While existing defenses show promise, they struggle to maintain robust security while preserving task functionality. We propose a novel and orthogonal perspective that reframes agent security from preventing harmful actions to ensuring task alignment, requiring every agent action to serve user objectives. Based on this insight, we develop Task Shield, a test-time defense mechanism that systematically verifies whether each instruction and tool call contributes to user-specified goals. Through experiments on the AgentDojo benchmark, we demonstrate that Task Shield reduces attack success rates (2.07%) while maintaining high task utility (69.79%) on GPT-4o, significantly outperforming existing defenses in various real-world scenarios.

Large Language Models (LLMs) are increasingly used for citation retrieval, yet their bibliographic outputs often contain hallucinated or inconsistent metadata. This paper examines whether structured prompting improves citation reliability compared with traditional API-based retrieval. We implement a three-stage BibTeX-fetching pipeline: a baseline Crossref resolver, a standard GPT prompting method, and a customized verification-guided GPT configuration. Across heterogeneous reference inputs, we evaluate retrieval coverage, field completeness, and metadata accuracy against Crossref ground truth. Results show that prompting improves coverage and completeness. Our findings highlight the importance of prompt design for building reliable, LLM-driven bibliographic retrieval systems.