Session-based Social Recommendation (SSR) aims to enhance next-item predictions by integrating a user's session activities with their social networks, yet it struggles with data sparsity, particularly due to the very limited short-term interactions within sessions.Traditional studies have applied Contrastive Learning (CL) to address this problem, but they typically avoid explicit data augmentations to keep original session sequences intact, considering the sensitivity of short sessions to disturbances.This conservative strategy, however, may limit the diversification of user-item interactions, potentially causing overfitting and generalization problems, and thus inadequately addressing data sparsity in session contexts.In this paper, we propose Intent Augmentation Contrastive Learning (IACL) for SSR, a model-agnostic approach that utilizes dual training objectives for sophisticated intent modeling.It starts with explicit intra-session augmentation to target specific intents and then expands to capture broader high-level intent patterns across sessions with augmented inter-session prototypes.In contrast to traditional strategies that skip augmentation, the core of IACL is its intent augmentation strategy, which offers both intra-session and inter-session perspectives, significantly broadening the model's exposure to various user behaviors and preferences, thus more effectively addressing data sparsity.Moreover, we introduce a denoised encoder module, Personalized Session Encoder with Filter Networks (PFN), which delivers personalized recommendations by incorporating all key features while suppressing noise in the frequency domain.By integrating PFN with IACL, we demonstrate that our approach, PFN-IACL, surpasses existing methods in performance across three datasets.Our code is freely available\footnote{Find our code at \url{https://anonymous.4open.science/r/PFN-IACL-WSDM26-32B5}}.