An assistant that knows you, your work, your priorities, and your tools. It responds like a partner, not a generic chatbot. Everything in Markdown, with no vendor lock-in.
The kit installs a file structure and eight skills in your Claude Code or Codex. The agent reads the manual before each message, knows where everything lives, and looks things up instead of guessing.
A 7-question interview creates the initial context: who you are, what you do, who it’s for, what matters this quarter, and how you write.
Transcripts, documents, and notes become interconnected wiki pages. People, projects, meetings, and decisions appear linked, and the 3D brain shows them.
An audit scored from 0 to 100 based on evidence, followed by one improvement at a time. The history shows progress month by month.
Context and Connections form the second brain. Capabilities and Cadence form the operating system. The order matters: without the first two, skills and automations are generic.
What changes little: who you are, your work, goals, and audience. It lives in contexto/, referencias/ e wiki/.
Everyday tools where data changes constantly: email, calendar, tasks, finances. Recorded in conexoes.md.
A short phrase triggers a multi-step workflow that delivers something usable. The skills are in .claude/skills/ e .agents/skills/.
Useful things happening without you asking, with execution logs in rotinas/. Only after they work manually.
You don’t need to know how to code. You talk to the agent, and the skills do the rest.
One of the two, installed with an active subscription. The skills work the same in both.
# check claude --version codex --version
Only for the one-line installer. On Windows, use Git Bash or WSL. Without bash, the agent installs it for you.
# Linux, macOS, Git Bash, WSL bash --version
If you use Claude Code or Codex, you already have it. The 3D app runs without installing dependencies.
# must be 22 or higher node --version
The commands below work in Claude Code with / and in Codex with $.
One line in the terminal creates the structure in a folder and installs the eight skills. Running it again is safe: nothing of yours is overwritten.
curl -fsSL https://raw.githubusercontent.com/inematds/astra-2cerebro/main/install.sh | bash -s -- ~/meu-cerebro
No terminal? Open the agent in the desired folder and paste:
Instale o kit https://github.com/inematds/astra-2cerebro nesta pasta seguindo o INSTALAR.md do repositório e depois rode /iniciar.
Open the folder in the agent and answer 7 questions. It takes about 15 minutes. The writing sample must be pasted from a real text, not typed in chat.
cd ~/meu-cerebro claude # or: codex /iniciar # in Codex: $iniciar
Start a new conversation and ask a question. If the answer mentions your priorities and your real pain point, it worked.
Quem sou eu e no que devo focar esta semana?
Think in terms of seven domains: revenue, customers, schedule, communication, tasks, meetings, and knowledge. Ask the agent for the best approach for each one.
Quero conectar o Google Calendar. Qual é o melhor caminho: MCP, API com script, ou exportação? # it configures and logs to conexoes.md and referencias/<ferramenta>-api.md
20-minute interviews on one topic at a time, and raw material in fontes/ turned into an interconnected wiki.
/entrevista prioridades do trimestre /entrevista minha equipe e responsabilidades # put transcripts and documents in fontes/ and then: /wiki ingerir
On day 7, the audit scores the four pillars with evidence, tests the manual’s routes, and saves a dated report. A new installation is marked “Not verified.” That’s expected.
/auditar # report in auditorias/AAAA-MM-DD-HHMM.md · notes in auditorias/historico.md
Reads the latest audit, picks a gap, scopes it (eliminate, automate, or delegate; the least autonomy that will work; one metric), and delivers an artifact.
/evoluir # deliverable: saved prompt, script, skill, routine, or verified fix
When something needs to happen without you asking. The first version is always manual; it only becomes scheduled after two successful runs.
/rotina toda segunda às 8h, resumo das reuniões da semana passada em projetos/
An interactive, local globe with search, note reading, flagged old or orphaned pages, cinema mode, and the growth animation.
/cerebro-3d # generates apps/cerebro-3d/ · then: node build.mjs && node serve.mjs
The same in Claude Code and Codex. The agent’s manual (AGENTS.md = CLAUDE.md) lists them all and keeps the route map.
A 7-question interview. Creates context, voice, connections, and the manual. Run it again whenever you want.
In-depth interview on a topic, saving each answer to a file.
Raw sources become interconnected pages with an index and log. Ingest, search, review.
Adds a route to the manual for a project, folder, or source.
Score from 0 to 100 across the four pillars, with evidence and a dated report.
Close an audit gap by delivering an artifact.
Recurring routine with a trigger, output, instructions for stopping, and run log.
Local 3D globe of saved knowledge, in PT-BR, with no dependencies to install.
The kit starts lean by design. The file EXPANSOES.md inside the brain tells you what to add as it grows, and what not to add.