Edge information is crucial in medical image segmentation tasks with complex boundaries, such as skin lesion segmentation, where precise edge modeling directly impacts segmentation accuracy. While Denoising Diffusion Probabilistic Models (DDPMs) have shown great potential in medical image segmentation due to their fine-grained detail refinement, existing methods generally rely only on the global features of the raw image as conditional input, failing to explicitly incorporate pixel-level edge priors during denoising. This limitation restricts segmentation accuracy, particularly in low-contrast transition regions. To address this, we propose EdgeSegDiff, a diffusion model for skin lesion segmentation based on an edge-conditional guidance strategy. The model introduces edge maps as conditional information and designs a joint loss function to dynamically regulate the dependence on edge priors. Additionally, an Edge Feature Enhancement (EFE) module is introduced, combining feature differencing and the SimAM attention mechanism to enhance edge feature representation. To effectively suppress high-frequency noise interference in the conditional input while retaining critical edge details, we further propose a High-Frequency Noise Filter (HF-Filter) based on wavelet transform and dynamic thresholding. On the publicly available ISIC-2016, ISIC-2017, and ISIC-2018 datasets, EdgeSegDiff achieves Dice coefficients of 92.1%, 87.4%, and 90.8%, respectively, demonstrating its superiority and robustness in skin lesion segmentation tasks.