My Claude Workflow

Decided to try to sketch out in the form of a diagram the current sequence of actions (the so-called Workflow) that make Claude work more effectively in order to implement my ideas.

Initially there is only code (or a repository with code). A CLAUDE.md file is created in the code folder, which Claude Code reads first. This is the starting point for him. Since this file is read every time, it should be as short as possible. Essentially it consists of links to other documents.

I also prefer to describe my thoughts in other applications (Visual Studio Code, Notes on iPad and so on). Based on this I create INPUT.md, which I either fill in or completely erase. After I have written all the notes and ideas, I simply tell Claude to update the documents, prioritize tasks and so on.

There are many entities in the project and I don't want to mix them up. I think it's logical that information about interface design should be separate from information about the codebase, or from business decisions. For this, a minimum of 4 files are used.

This is essentially an analogue of a Kanban board or something like that. As if I myself in a task tracker was moving tasks back and forth. Initially everything goes into the backlog. Several tasks that need to be done are moved to the sprint. After successful completion of a task, the status is updated in the sprint and in the change history.

The Justifier is a block that plays the role of a sage. He can always say who, what, when, how and why did what. And if an error occurred, he knows how to deal with it. Otherwise he tries to fix it and writes the information to the corresponding file.

Kind of like that. The diagram is far from ideal and I didn't really think about it for long, but minimal control and double-checking itself does perform.