Abstract: Precise monitoring and quantitative inversion of coal–rock fracture geometries and stress‐field distributions are regarded as fundamental scientific prerequisites for elucidating the evolution of rockburst processes and associated failure mechanisms, and are considered an essential approach for resolving dynamic disaster evolution in deep engineering. Microcrack source models corresponding to different fracture types are established on the basis of the displacement discontinuity theory. By modifying the Bott assumption, a stress inversion model applicable to tensile–shear fractures is proposed, and a composite failure criterion is constructed through the integration of the Griffith and Mohr–Coulomb criteria, enabling effective identification of true fracture planes. With the optimal compatibility between source‐mechanism parameters and stress‐field parameters taken as the target, a joint iterative inversion method of microcrack sources and stress fields is further developed, followed by laboratory‐scale monitoring and inversion experiments on coal–rock failure. The results indicate that the offset angle α effectively discriminates tensile, shear, and compressive fracture types. The iterative solution method for fracture planes and stress based on the tensile–shear composite failure criterion enables robust selection of true fracture planes and improves stress inversion accuracy. Establishment of a joint alternating inversion model of microcrack sources and stress effectively reduces the deviation between theoretical and actual slip directions. During three‐point bending tests on coal–rock specimens, α predominantly ranges between 0° and 20°, indicating that microcracks mainly exhibit tensile–shear composite characteristics. The stress inversion model yields a maximum principal stress σ₁ in tension, whereas σ₂ and σ₃ are compressive. The average stress shape ratio reaches 0.87, demonstrating the dominance of σ₁ and the near equivalence of σ₂ and σ₃, consistent with the loading state of three‐point bending specimens. The stress distribution governs crack motion predominantly as tensile separation along the X-axis and promotes crack development preferentially oriented along the Y-axis. Moreover, stress rotation reduces the Y-axis component of fracture‐plane orientation, increases the angle between crack motion and the X-axis, and this alteration intensifies with increasing rotation angle. Optimization of the stress inversion process using the joint inversion method progressively lowers the mean slip deviation angle between theoretical and actual slip directions by 33.91%, 79.86%, and 94.12%, respectively. Compared with the VAVRYČUK method, the proposed tensile–shear source stress inversion method achieves up to a 6.49% reduction in the noise error rate of the stress shape ratio, while improving the average accuracy of three‐directional stress inversion by approximately 15%–40%. These findings provide a rigorous methodological framework for investigating the dynamic failure evolution of coal–rock masses and offer new theoretical support for deciphering the incubation mechanisms of rockburst hazards and advancing disaster early-warning strategies in deep engineering.