Do you maintain a prompt log or is it covered by your commit history?
By week three of a side project, I could not recall why I had programmed the application to deal with time zones in that particular manner. The code made sense within itself, but the rationale was missing because it existed in a chat window that I had long since closed.
Hence, I began creating a simple text file in the repository entitled "decisions," and in it, I noted two or three sentences: the question, the decision I made in response, and the date. This took less than a minute, but it has spared me from reversing my decision twice.
The log is different from a comment within the code. The comment tells me what the code does. The log tells me why we took a different route because this is the information I tend to forget.
Are you keeping any similar records or are your commit notes doing the trick? I would like to hear whether there is a better way of keeping a record, perhaps something universal to both Cursor and Claude Code.
Replies
Same conclusion here, commit messages were not enough. Two things I would add to your decisions file after a couple of months of doing it. Split it by area once it grows: I have one notes file per part of the app, and the agent reads the matching one before it touches that part, so the old reason actually reaches the moment it matters. And record the things that failed, not only the choices. A line like "the yearly plan was preselected until October 5, 15 of 16 people who tapped Continue cancelled on the store sheet" stops me and the agent from trying it again in three months. The dated why is the useful part, exactly as you say.