Does anybody else keep a decision journal for his/her AI agent?
by•
But already in two weeks into working with Plantlog by Claude Code, I saw that every session was offered to use a different state library by the software without recalling our choice of Zustand.
Therefore, I decided to create an ordinary markdown file called DECISIONS.md, which is to have a single line of information about each of our decisions and another single line of reasoning for that. I copy and paste that file every time we start a new session.
The issue of suggestions is now solved. Does anybody know some alternative way to do that?
8 views
Replies
You can skip the copy-paste. Claude Code reads a CLAUDE.md file from the project root at the start of every session, so if you rename DECISIONS.md to that, the Zustand line is already loaded before you type anything.
My project file works the way yours does: 1-line per rule, and the reasoning lives in a separate doc the agent only opens when it touches that area. Once a line grows into a paragraph, move the paragraph out. A long file gets skimmed by the agent the same way it gets skimmed by people.
! However, where I don't trust the file alone: decisions that would quietly hurt if they slipped. For example, I retired some naming in my codebase, and a test fails if the old names come back. The journal tells the agent what we decided. The test is what stops it