This paper presents the design, analysis, and performance evaluation of an attitude filter for an Intervention- Autonomous Underwater Vehicle (I-AUV) working in tandem with an Autonomous Surface Craft (ASC). In the proposed framework, an ASC aids an I-AUV to determine its attitude by providing an inertial reference direction vector through acoustic modem communications, which is measured in body-fixed coordinates by an Ultra-Short Baseline (USBL) device. The specificity of the intervention mission to be carried out, near structures that distort the Earth magnetic field, renders on-board magnetometers unusable for attitude determination. The solution presented herein includes the estimation of rate gyros biases, yielding globally asymptotically stable error dynamics under some mild restrictions on the vehicle team configurations. The feasibility and performance of the proposed architecture is assessed resorting to numerical Monte-Carlo simulations with realistic sensor noise, and transmission delays and limited bandwidth of the acoustic modems.