Open Source Software(OSS) is crucial in modern software ecosystems, with README files serving as key sources of information regarding software functionality, configuration, and usage instructions. However, approximately 9.03% of OSS projects suffer from missing, incomplete, or poorly structured README files, hindering developer efficiency and software maintainability. Recent studies have utilized natural language processing and large language models(LLMs) techniques to automate README generation, aiming to reduce manual documentation efforts. Despite notable progress, existing methods face challenges such as integrating real-time external knowledge, parsing complex software structures, and inadequate customization for project-specific requirements. To address the aforementioned challenges, this paper proposes RMGenie, a framework leveraging the advanced comprehension capabilities of LLM-based agents and their ability to invoke external tools to automate the generation of OSS README documents. First, RMGenie constructs an automated agent workflow to facilitate multi-round interactions with external tools, dynamically integrating external knowledge to progressively extract critical code information for complex software structures. Second, an action tree model is established to quantify the decision paths taken by the agent, employing an information entropy-based scoring mechanism to select the optimal candidate path, thereby determining the most effective README generation decision sequence. Finally, to mitigate issues such as decision biases and abnormalities in tool invocation within the workflow, RMGenie incorporates a feedback mechanism that enhances the accuracy and success rate of the generation process. Experimental results demonstrate that RMGenie significantly outperforms existing baseline methods in terms of content completeness, instruction adherence, and factual accuracy. Additionally, we develope OSS Readme Generator, a VSCode plugin leveraging RMGenie to automatically generate README files by retrieving relevant information from GitHub repositories, thereby enhancing developer productivity and improving OSS documentation standardization.