Claude Code Directory: Settings, Files & Folder Structure
Where are Claude settings stored? A plain tour of the .claude directory files: CLAUDE.md, settings.json, skills, memory — and what is safe to touch.
Evgenii Arsentev · PhDReviewed for accuracy by Evgenii Arsentev, PhD · 2026-06-12
Claude Code keeps its settings and files in two places: a .claude directory inside your project, and a ~/.claude directory in your home folder for everything personal. Once you know what lives where, the whole "mysterious hidden folder" anxiety evaporates.
I'm Evgeny, and I remember the day I first noticed a .claude folder had appeared in my project and wondered if I was allowed to look inside. You are. In fact, peeking inside is the single fastest way to understand how Claude Code actually thinks about your project. Let's take the tour, room by room.
Where are Claude settings stored?
In layered settings files, and the layers have a strict pecking order. From strongest to weakest: managed policy from your organization, command-line flags, your personal overrides in .claude/settings.local.json, the project's shared .claude/settings.json, and finally your global defaults in ~/.claude/settings.json. A value set higher up wins over the same value lower down. Translation for humans: your local file beats the team file, and the team file beats your global defaults.
How do I find the Claude directory?
Both directories start with a dot, which makes them invisible in most file browsers — that's a convention for "configuration, not content", not a secrecy plot. The project one sits at the root of any folder where you've used Claude Code; the personal one is always at ~/.claude in your home directory.
ls -a ~/.claude
The -a flag shows dot-files. Run `ls -a` inside a project to spot its local .claude/ too. On Windows, check your user folder for the same names.
Where does Claude store my files and memory?
Conversations and memory live under ~/.claude/projects/, organized per project. Inside each project's memory folder, Claude maintains a MEMORY.md file all by itself — notes it takes when you correct it or state preferences. The first 200 lines load at the start of every session, which is why Claude "remembers" that you hate emojis. This auto memory is the mirror twin of CLAUDE.md: CLAUDE.md is what you write for Claude, MEMORY.md is what Claude writes for itself.
What are all those other folders — rules, skills, agents?
Each subfolder is an extension point, and each exists in both territories: project-level for the team, ~/.claude for just you. rules/ holds topic-scoped instructions that can load only when relevant files are touched. skills/ holds reusable workflows you invoke as /name. agents/ defines subagents with their own separate memory.
What about the commands folder?
commands/ is the older single-file version of skills — old files still work, but new workflows belong in skills/. None of these folders are required; an empty .claude folder is a perfectly healthy .claude folder.
How do I change the directory for Claude?
You don't configure it anywhere — Claude Code simply works in whatever folder you launch it from, and each session is tied to that directory. Want Claude on a different project? Just cd into that folder first, then run `claude`. That's the entire mechanism, and it's also why running Claude from your home directory is messy: it sees everything you own.
Lab: leave your first note for Claude
0/5Safe to edit: CLAUDE.md, anything in rules/ and skills/, and settings.local.json — it's yours alone. Be careful with the team's settings.json (others rely on it), and never put passwords or API keys in any committed file — settings.json travels with the project, settings.local.json stays on your machine. That split exists precisely so secrets and personal quirks don't leak into shared code.
Tonight's homework: open ~/.claude/projects/ and read your own MEMORY.md files. You'll see exactly what Claude has learned about how you like to work — and once you realize the whole system is just readable text files in folders, you'll stop fearing configuration forever. It's not a black box. It's a filing cabinet, and you've got the key.
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →References
Selected peer-reviewed papers and preprints on the AI research behind this topic.
- [1]Shrivastava et al. (2022). Repository-Level Prompt Generation for Large Language Models of Code. arXiv:2206.12839
- [2]Wu et al. (2024). Repoformer: Selective Retrieval for Repository-Level Code Completion. arXiv:2403.10059
- [3]Liu et al. (2024). STALL+: Boosting LLM-based Repository-level Code Completion with Static Analysis. arXiv:2406.10018
- [4]Deng et al. (2024). R2C2-Coder: Enhancing and Benchmarking Real-world Repository-level Code Completion Abilities of Code Large Language Models. arXiv:2406.01359
- [5]Liu et al. (2024). M2rc-Eval: Massively Multilingual Repository-level Code Completion Evaluation. arXiv:2410.21157
- [6]Zhang et al. (2025). CodeRAG: Finding Relevant and Necessary Knowledge for Retrieval-Augmented Repository-Level Code Completion. arXiv:2509.16112
- [7]Khan et al. (2025). Kodezi Chronos: A Debugging-First Language Model for Repository-Scale Code Understanding. arXiv:2507.12482
Author
Evgenii Arsentev
PhD · Chief Executive Officer, digital health
Reading is a good start
Want to actually build this?
Guides explain. The free course transforms — personalized, gamified, and built to get you shipping fast.
Start the free course →