A ready-made "AI operating system" for Claude Code: a folder that makes Claude act like a personal assistant, study partner and build partner that knows who you are, follows your processes, and remembers things between sessions.
Built for a university student who is new to all of this, for two jobs:
- University: lecture notes, quizzing, assignment coaching, research
- Building custom AI agents for your own needs
What you get:
| 17 skills | notes, quizzing, assessment coaching, research, spec/build/review/debug, agent and skill builders, memory, session handoff |
| 6 agents | researcher, tutor, marker, code-reviewer, debugger, explainer |
| 6 rules | how to talk to you, verify before done, security, git safety, academic integrity, how to brief agents |
| G-Brain | long-term searchable memory, local and free |
| Guides | docs/: how it works, models and effort, G-Brain, Orca, vibe coding, building agents, cheatsheet, glossary |
- A free account at github.com.
- A paid Claude plan (Pro or Max). The free plan does not include Claude Code.
Open a terminal (Mac: press Cmd+Space, type "Terminal". Windows: open "PowerShell") and paste:
# Mac or Linux
curl -fsSL https://claude.ai/install.sh | bash# Windows (PowerShell)
irm https://claude.ai/install.ps1 | iexWindows users: also install Git for Windows.
Close the terminal, open a new one, and check it worked: claude --version.
Then run claude once and log in through the browser. Type /exit to leave.
Orca is the desktop app you will run Claude in. See docs/orca.md.
Mac: brew install --cask stablyai/orca/orca, or download it from
github.com/stablyai/orca.
You can skip Orca and just use claude in a terminal; everything works the
same.
Your copy will hold personal things (units, notes, assignments), so it must be private. Claude can make it for you.
In a terminal, go to your home folder and start Claude (or start an agent in Orca in your home folder):
cd ~
claudeThen paste the prompt from docs/first-prompt.md.
It installs nothing without asking, gets you logged in to GitHub, creates a
private repo called my-ai-os from this template, downloads it, sets up git,
and proves each step worked.
Prefer to do it by hand?
- On this repo's GitHub page click Use this template, then Create a
new repository, name it
my-ai-os, choose Private. - Install the GitHub CLI (cli.github.com) and run
gh auth login(GitHub.com, HTTPS, web browser). cd ~ && git clone https://github.com/<your-username>/my-ai-os.gitgit config --global user.name "Your Name"andgit config --global user.email "you@example.com"(your GitHub email).
Type /exit, then:
cd ~/my-ai-os
claude(Orca: add the my-ai-os folder as a project and start a Claude Code agent
in it.) Your skills only load when Claude starts inside this folder. When it
asks whether you trust the folder, say yes.
The first-run skill interviews you (about 10 questions) and fills in
context/ with who you are, your units, deadlines and goals.
The gbrain-setup skill installs and connects your memory layer, step by
step (about 10 minutes). Read docs/gbrain.md for what it is.
bash scripts/check-setup.shWindows: run this in Git Bash (search "Git Bash" in the Start menu; it
came with Git for Windows), from inside your my-ai-os folder.
Everything should say [ok]. If not, paste the output to Claude.
Then read docs/how-it-works.md. Ten minutes, and everything else makes sense.
You just talk to it. Some things to try first:
| Say | What happens |
|---|---|
drop slides in uni/_inbox/, then "make notes" |
study notes in uni/units/<CODE>/notes/ |
| "quiz me on week 3" | active-recall quiz, tracks weak topics |
| "help me with my assignment" | decode the brief and rubric, plan, research, critique |
| "mark my draft" | rubric-based marking with the top 3 fixes |
| "explain X like I'm new" | the tutor or explainer agent |
| "research X" | cited research summary |
| "build me an agent that..." | spec, build, review, step by step |
| "remember that..." | saved to G-Brain |
| "wrap up" | session summary, status update, commit, memory sync |
Full list: docs/cheatsheet.md.
CLAUDE.md the brain file Claude reads every session
context/ who you are: me, goals, priorities, study, connections
hot-cache/ what is happening this week
uni/ units, notes, assessments, writing samples
second-brain/ everything else you learn, as a linked wiki (opens in Obsidian)
projects/ things you build
specs/ plans written before building
decisions/ decision log
reminders/ future deadlines
inbox.md dump anything here
templates/ fill-in forms
docs/ the guides
.claude/rules/ standing rules
.claude/skills/ processes
.claude/agents/ specialist helpers
Details: docs/how-it-works.md.
This is a starting point, not a finished product. It gets better the more you shape it:
- The same request three times: say "make that a skill".
- The same mistake twice: say "add a rule so that doesn't happen again".
- A job you want handed off: say "make an agent for that".
- Every skill ends with a self-improvement check that proposes a fix to itself when something went wrong, and never edits without your yes.
Check your university's and each unit's AI-use policy and record it in
context/study.md. The system coaches by default (understanding, planning,
sources, feedback) rather than writing your graded work, and it never
invents sources. You own everything you submit.
Structure and methods adapted from a personal Claude Code operating system in daily use. Memory layer: G-Brain by Garry Tan. Agent environment: Orca by Stably AI. The "grill me" interview pattern is inspired by the public grill-me skill.
MIT licensed. See LICENSE.