Reconstructing a complete 3D scene from a single panoramic image is a critical task for computer vision and graphics. Existing methods often fail to recover unobserved regions with geometric and semantic consistency, which limits the quality of novel view synthesis. To address these challenges, we present PanoExplorer, a novel framework that reconstructs high-fidelity indoor scenes. Our method adaptively constructs a camera trajectory to explore occluded areas and completes unobserved content using a view-conditioned inpainter guided by pose-aware textual descriptions. The process begins with an initial mesh generated from the input panorama, which is iteratively refined by inpainting missing content from synthesized viewpoints. This refined mesh is then transformed into a 3D Gaussian Splatting (3DGS) field and optimized. To enhance scene fidelity, we introduce a customized Gaussian upsampling strategy that improves geometric continuity and preserves texture details, particularly at object boundaries. Extensive experiments demonstrate that PanoExplorer reconstructs high-quality 3D indoor scenes from a single panorama in approximately 35 min, significantly outperforming state-of-the-art methods. Our approach provides a robust foundation for downstream tasks such as immersive walkthrough and virtual scene modeling for AR/VR applications.