AURA: Intent-Directed Probing for Implicit-Need Surfacing in Situated LLM Agents

In three linesAURA is a situated LLM agent system that detects implicit needs beyond literal query content. Between scene perception and tool use, it generates a structured IntentFrame with a gap score to control probing budget. On a 100-query benchmark, AURA improves implicit-need coverage by +0.07 (p < 10^-6) versus ReAct-style probing.

Summary generated by Claude — human-verified