AIOS Starter Kit
GerardoRdz96/aios-starter-kit
Sistema operativo de IA pessoal que aprende e cresce com você
Ver no GitHubTemplate clonable para construir um Sistema Operativo de IA pessoal que mantém contexto, conhecimento e skills entre sessões. Inclui onboarding interativo, wiki de conhecimento, comandos reutilizáveis e builders para criar novos agentes. Oferecido em versão bilíngue (EN/ES) com segurança integrada.
Instalação
npx convez-ai GerardoRdz96/aios-starter-kit1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Construir um assistente IA que mantém contexto e memória pessoal
- Automatizar workflows repetitivos com /commands customizados
- Criar e gerenciar uma wiki de conhecimento persistente entre sessões
Veredito

Template bem estruturado e maduro (v2.1) que resolve o problema real de IA sem memória. Abordagem hybrid EN/ES com segurança declarada é rara. Ideal para quem quer mais que chat — um sistema que evolui.
README
A tradução para o português deste README ainda está na fila.
🧠 AIOS Starter Kit
Build your own personal AI Operating System.
A clonable template that turns Claude Code into a thought partner that remembers — it knows who you are, compounds knowledge across sessions, and grows new skills over time.
Un sistema operativo de IA personal. Una plantilla para clonar que convierte a Claude Code en un compañero de trabajo que recuerda.
Kit v2.1 — June 2026
What is an AIOS? / ¿Qué es un AIOS?
A normal AI chat forgets everything when you close it. An AI Operating System doesn't.
It's a folder — version-controlled, owned by you — that gives your AI assistant four things a blank chat never has:
- Context — it knows who you are and what you're working on.
- A knowledge wiki — facts it has learned, written down and reused (not lost in chat history).
- Skills — repeatable workflows you trigger with
/commands. - A way to grow — builders that let it create new skills, agents, and automations.
You don't program it. You talk to it, and it writes itself into a system that gets more useful every week.
Español: Un chat normal de IA olvida todo cuando lo cierras. Un Sistema Operativo de IA (AIOS) no. Es una carpeta — con control de versiones, tuya — que le da a tu asistente cuatro cosas que un chat en blanco nunca tiene: contexto (sabe quién eres), una wiki de conocimiento (lo que ha aprendido, escrito y reutilizable), skills (flujos de trabajo que activas con
/comandos), y una forma de crecer (constructores que crean nuevas skills, agentes y automatizaciones). No lo programas: hablas con él.
Quickstart / Inicio rápido
You need Claude Code installed.
# 1. Clone the kit / Clona el kit
git clone https://github.com/GerardoRdz96/aios-starter-kit.git my-aios
cd my-aios
# 2. (optional) make it your own private repo / hazlo tu propio repo privado
rm -rf .git && git init
# 3. Arm the security guards / Activa los guardias de seguridad
scripts/rails-guard.sh install
# 4. Open Claude Code in the folder / Abre Claude Code en la carpeta
claude
# 5. Inside Claude Code, run the onboarding wizard / Corre el asistente de bienvenida
/onboard
/onboard interviews you (about 7 short questions), then fills in your context. That's it —
you have a working AIOS. Try asking it: "what should I focus on this week?"
Security: your personal dirs (context/, knowledge/, decisions/, artifacts/) are gitignored by default, and the guards from step 3 block accidental pushes of personal content + silent edits to governance files. Threat model and the rules: SECURITY.md.
Español: Necesitas Claude Code instalado. Clona el kit, ábrelo con
claude, y corre/onboard. Te hará unas 7 preguntas y llenará tu contexto. Listo: ya tienes un AIOS funcionando. Pregúntale: "¿en qué me debo enfocar esta semana?"
What's inside / Qué incluye
my-aios/
├── CLAUDE.md ← the operating system: who your AI is + how it works
├── AGENTS.md / GEMINI.md ← entry points for non-Claude tools (Codex, Gemini, Cursor…)
├── aios-intake.md ← the 7-question intake /onboard reads
├── pending.md ← open follow-ups that must survive between sessions
├── .claude/
│ ├── skills/ ← /commands you run (rituals + builders + multi-brain, 13 total)
│ ├── agents/ ← specialized helpers (scribe, warden)
│ └── teams/ ← multi-agent crews you build later
├── scripts/ ← deterministic helpers (e.g. the weekly scoreboard)
├── context/ ← what your AI knows about YOU (filled by /onboard)
├── references/ ← your knowledge wiki (the AI owns and writes this)
├── knowledge/ ← raw sources you drop in (transcripts, PDFs, notes)
├── artifacts/ ← deliverables your AI builds for you
├── decisions/ ← append-only log of decisions + why
├── routines/ ← scheduled / recurring automations
└── archives/ ← old stuff (move here, don't delete)
The skills it ships with / Las skills incluidas
| Command | What it does |
|---|---|
/onboard | Day-1 setup wizard. ~7 questions → your context files. |
/audit | Health check. Scores your AIOS on the Four Cs (Context, Connections, Capabilities, Cadence). |
/level-up | Weekly ritual. Find one thing to automate, scope it, ship it. Runs on the 3Ms (Mindset, Method, Machine). |
/grill-me | Deep extraction interview. One question at a time, checkpointed to disk, graduates into your wiki. |
/session-handoff | Wrap-up note (what we did, open decisions, next steps) so a fresh session — in any tool — continues cold. |
/skill-builder | Build a new skill. |
/agent-builder | Build a new specialized agent. |
/routines-builder | Build a scheduled / recurring automation. |
/agents-team-builder | Design a small team of agents that work in parallel. |
/workflow-builder | Build a saved dynamic workflow — N parallel sub-agents fanning out over a big task. |
/hooks-builder | Build an event-driven hook ("every time X happens, do Y") with a mandatory supervised test. |
/plugin-builder | Package your skills/agents into a shippable, installable plugin. |
multi-brain | Auto-router for all your models — cloud CLIs and local ones. You define the roster; it enforces the No-Self-Review Law (a model never reviews its own work). |
The first five are rituals (you run them on a cadence or a moment). The next seven are
builders — every capability type your AIOS can have now has a builder, which means it
can grow itself. multi-brain has no slash command; it fires on its own when a route
matters. Before you let any of it run unattended, read
references/autonomous-entity-charter.md —
the governance layer: the AI builds, you arm.
Power skills (install once, user-global)
The kit assumes a small stack of public power skills that work in every project once
installed: Superpowers (plan → test-first → self-review), GSD (fresh sub-agent per
task for long builds), Frontend Design (anything visual), the built-in /review, and
the memory pair Context Mode + Claude Mem (auto via hooks: leaner context + cross-session
memory). Install commands + the route-by-risk doctrine:
references/power-skills.md.
The big idea: a knowledge wiki that compounds
The most important folder is references/ — your AIOS's knowledge wiki, built on
Andrej Karpathy's LLM Wiki pattern. Three layers:
knowledge/ → references/ → CLAUDE.md + wiki-protocol.md
(raw sources) (the wiki the (the rules that keep
you drop in) AI writes & owns) the wiki disciplined)
You drop a source in knowledge/. Your AI reads it, discusses it with you, and writes an
interpreted page into references/. Next session, it reads that page instead of starting
cold. Knowledge compounds instead of resetting every chat. The rules live in
references/wiki-protocol.md.
Español: La carpeta más importante es
references/— la wiki de conocimiento de tu AIOS, basada en el patrón LLM Wiki de Andrej Karpathy. Tú sueltas una fuente enknowledge/, la IA la lee, la discute contigo, y escribe una página interpretada enreferences/. La próxima sesión lee esa página en vez de empezar de cero. El conocimiento se acumula en vez de reiniciarse en cada chat.
How it grows / Cómo crece
Once you're onboarded, the loop is simple:
- Run
/auditon day 7 and then weekly — it scores your setup and names the top gaps. - Run
/level-upweekly — it finds one manual thing you keep doing and helps you turn it into a skill, an agent, or a routine. - Drop sources into
knowledge/as you learn — the wiki grows. - Build with the seven builder skills whenever you hit a repeating need.
- Govern the growth: the autonomous-entity charter
sets the risk tiers, the measurement scoreboard (
scripts/entity-scoreboard.py), and the five human-in-the-loop gates that keep a self-building AIOS honest.
Your AIOS in month three looks nothing like day one — because you grew it.
Works beyond Claude Code / Funciona más allá de Claude Code
An AIOS is folders and markdown — nothing here is locked to one vendor.
- Claude Code CLI — the native experience: skills load as
/commands, agents fork automatically. - Claude Desktop (Cowork tab) — point Cowork at your AIOS folder and talk. Same brain, same wiki, no terminal. Great for non-coders; switch to the CLI when you want skills as slash commands.
- Codex CLI / Gemini CLI / Cursor / others —
AGENTS.mdandGEMINI.mdroute any harness toCLAUDE.md, the single operating manual. Skills are written as plain procedures, so any agent can follow them.
And with the multi-brain skill, your models stop being either/or: one assistant drives, and it routes sub-tasks to whichever brain — cloud or local — fits best.
Español: Un AIOS son carpetas y markdown — nada está amarrado a un solo proveedor. Funciona en Claude Code, en la pestaña Cowork de Claude Desktop (apunta Cowork a tu carpeta y conversa), y cualquier otro agente entra por
AGENTS.md. Conmulti-brain, tus modelos dejan de ser "uno u otro": uno maneja y enruta al que mejor le quede.
A note on privacy / Una nota sobre privacidad
This kit ships empty of personal data. Once /onboard fills your context/ files, that
folder is about you. If you make your copy public, check .gitignore — there are
commented lines to keep context/, knowledge/, and decisions/ private. Never commit
secrets (.env files are already ignored).
Credits / Créditos
- The knowledge wiki pattern: Andrej Karpathy's LLM Wiki idea.
- The 3Ms framework (Mindset, Method, Machine): a framework by Nate Herk.
- Built on Claude Code by Anthropic.
Derived and generalized from a personal AIOS. Shared so you can build your own.
License: MIT — see LICENSE. Use it, fork it, make it yours.