A Readme file is usually a text file titled README.txt that should be located at the root of your dataset. Its title indicates that any potential user of your data should consult it before checking any other part of your dataset.
It is a form of documentation that can be created manually, without using a specific standard, and is therefore often easier to read. Your readme can also take the form of a PDF/A document or use a different title, as long as it is clearly labeled as something the user should read. It is generally better to start writing up that information into a central document as soon as you start collecting your data, or even during the planning phase.
The main readme file explains the contents and structure of your dataset, and gives enough information for a potential user to determine whether the data is of interest to them or not. If your dataset requires a codebook, it can be included within it. You can of course also create secondary readme files in subfolders to document specific parts of your data.
The following is a suggested list of elements you could include in a readme file located at the root of your dataset. You may of course place some of them in secondary documentation or in separate files, or even embed them in your data. Many of the suggestions only make sense with certain types of data, and the goal is not to make your readme file too lengthy. You should only include elements that are useful and/or necessary to correctly interpret, evaluate and reuse your dataset.