Docker is a popular tool for developers and organizations to package, deploy, and run applications in a lightweight, portable container. One key component of Docker is the Dockerfile, a simple text file that specifies the steps needed to build a Docker image. While Dockerfiles are easy to create and use, creating an optimal image is complex in particular since it is easy to not follow the best practices, when it happens we call it Docker smell. To improve the quality of Dockerfiles, previous works have focused on detecting Docker smells, but they do not offer suggestions or repair the smells. In this paper, we propose, Parfum, a tool that detects and automatically repairs Docker smells while producing minimal patches. Parfum is based on a new Dockerfile AST parser called Dinghy. We evaluate the effectiveness of Parfum by analyzing and repairing a large set of Dockerfiles and comparing it against existing tools. We also measure the impact of the repair on the Docker image in terms of build failure and image size. Finally, we opened 35 pull requests to collect developers' feedback and ensure that the repairs and the smells are meaningful. Our results show that Parfum is able to repair 806 245 Docker smells and have a significant impact on the Docker image size, and finally, developers are welcoming the patches generated by Parfum while merging 20 pull requests.