What's in the Folder
Everyone who uses Claude Code has this folder sitting in their project, but few can explain the structure inside it. The article dissects it file by file: where config and permission rules go, how memory and instruction files like CLAUDE.md get loaded, the location of session state and caches, and which should go into version control and which into gitignore. This knowledge isn't usually taught—you find it's valuable only when something breaks, like a permission rule written in the wrong place not taking effect, or committing a local config with sensitive information into the repo.
Config Is Leverage
The article's implicit theme is: a considerable part of an agent tool's true capability ceiling is determined by these unremarkable config files. A good CLAUDE.md is equivalent to preloading project knowledge into every session, the tightness of permission rules directly determines the balance of security and smoothness, and once a team puts these files into version control, personal experience becomes a team asset. The community's "dotfiles culture" is extending into a ".claude culture," with people starting to share and compare their configs. The more autonomous the tool, the longer the leverage of these config files—spending half a day sorting them out is usually more worthwhile than switching to yet another new model.
via: Hacker News