Abstract: This paper presents an algorithm-driven framework for the automated design of a passive reconfigurable textile metasurface featuring dual broadband, polarization-insensitive absorption and reflection. Our approach overcomes the limitations of single-function devices and complex active designs. The methodology is a two-stage computational process. First, a physics-constrained evolutionary search optimizes a pixelated resistive film for broadband absorption. Second, a density-driven greedy layering algorithm post-processes the pattern, decoupling resonant cells to enable a low-loss reflection mode when paired with a reflecting surface. The resulting design achieves over 90% broadband absorption, insensitive to both TE and TM polarizations. When mechanically switched to reflection mode, the resistive film has less than 1 dB of insertion loss and provides a 3 dB gain enhancement for an integrated antenna. In conclusion, this automated framework offers an efficient solution for the inverse design of multifunctional metasurfaces. It surpasses traditional methods by discovering non-intuitive, high-performance structures, providing a new design tool for reconfigurable devices, smart skins, and stealth technology.