Educational Assessment Agents for Multidimensional Scenarios: Development, Application, and Effectiveness Validation

Competency-oriented assessment reform in basic education must overcome the traditional emphasis on knowledge over competencies and outcomes over processes. Large language model-driven educational assessment agents offer new possibilities for multidimensional content, precise feedback, and real-time assessment, yet their development and validation in authentic multidimensional scenarios remain limited. This study developed Co-Quiz for five scenarios: assessment content generation and evaluation, knowledge and skill diagnosis, competency assessment, classroom emotion analysis, and classroom behavior analysis. It designs a multi-level collaborative functional system, formulate a configuration and development scheme, and guide sustained teacher-student-agent interaction, and implemented Co-Quiz on a low-code platform. An application strategy was then proposed and evaluated through expert ratings, questionnaires, and semi-structured interviews. Results showed high expert ratings, relatively high teacher and student satisfaction, and clear advantages in assessment effectiveness and learning support. This study offers practical guidance for developing, applying, and evaluating educational assessment agents and advancing the intelligent transformation of educational assessment.