FME run process for each record separately instead of accumulating input records

I have multiple files under a directory. My objective is to perform a process on each file in the folder. This process basically reads all features of a file and generate a bbox.

I use a file and directory reader to get all the files' path and name. Then use them as initiator for a feature reader. My expectation is that the feature reader reading all features of one file each time and the following steps generating bbox for it and repeat the process. The final result would be multiple bboxes. However, the feature reader accumulates features from all of the files and the final result is one bbox.

Is there any way to handle this flow?