A MODULAR SOFTWARE FRAMEWORK FOR NEURAL-AUGMENTED SELF-MODELING AGENTS WITH EXPLICIT INTERNAL STATE REPRESENTATION

Self-modeling agents capable of explicitly representing their own internal cognitive states are critical for advancing adaptive and trustworthy artificial intelligence. This work introduces a modular software framework for neural-augmented self-modeling agents that dynamically monitor and regulate key internal variables such as confidence, fatigue, and behavioral mode. The architecture integrates dedicated neural submodels and a meta-learner module to enable real-time introspective adaptation across both single-agent and multi-agent configurations. Extensive experiments demonstrate that agents equipped with explicit internal state representation achieve stable meta-cognitive control, adaptive exploration-exploitation balance, and measurable self-awareness in dynamic environments. The fully engineered framework provides transparent instrumentation, reproducibility tools, and extensibility for broader research in meta-cognitive AI and adaptive multi-agent systems.