High-Frequency Trading (HFT) plays a crucial role in real cryptocurrency markets, where rapid decision-making is essential to capture short-term market opportunities. In parallel, social media platforms such as Reddit and Twitter have emerged as valuable sources of information that may impact investor behavior and market dynamics. While existing research has explored the use of social media data and Large Language Models (LLMs) in financial forecasting, their application in high-frequency, short-term decision-making remains underexplored. In this work, we introduce a social media dataset curated to facilitate the study of its influence on high-frequency trading strategies. Combined with publicly available cryptocurrency price data from Binance, we investigate the potential of LLM-based reasoning in the multi-agents framework to enhance short-term trading performance. Preliminary experiments under different market conditions suggest that incorporating social media information can provide marginal improvements in trading outcomes. Our dataset and findings offer a foundation for further exploration of MAS decision-making in high-frequency trading environments. We have open-sourced the dataset at \url{https://anonymous.4open.science/r/PulseReddit-ACE3/}.