convez.ai
SkillProdutividadeIntermediário
marciohideaki

Second Brain Starter

marciohideaki/second-brain-starter

Segundo cérebro com IA que realmente lembra o que você ensinou

Ver no GitHub
InicialPluginScore · de 0 a 100

Kit para construir um segundo cérebro alimentado por IA que persiste entre sessões. Fornece 12 skills, 4 hooks e 2 cron jobs para ingerir fontes, organizar conhecimento e compilar wikis pessoais em markdown. Suporta Claude Code, Cursor, Gemini CLI e outros agentes.

Instalação

$npx convez-ai marciohideaki/second-brain-starter

1 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

Estrelas
0
Forks
0
Acompanhando
0
Saúde
71%
Último commit
há 2 meses
Versão
v0.1.0
Issues abertas
0
Contribuidores
2

Casos de uso

  • Capturar e organizar aprendizados de leitura contínua
  • Manter contexto de projetos entre sessões
  • Construir wiki pessoal automaticamente indexada
  • Gerar briefings diários com síntese do conhecimento acumulado
  • Conectar ideias relacionadas em base de conhecimento persistente

Veredito

Solução robusta para quem trabalha com conhecimento e precisa que a IA retenha contexto além de uma sessão. Bem documentada em PT-BR, código limpo em Shell, e reutilizável como template.

README

A tradução para o português deste README ainda está na fila.

second-brain-starter

A starter kit to build your own AI-powered second brain using Claude Code — that actually remembers what you taught it, session after session.

License: MIT CI Claude Code

Who this is for: software developers, content creators, marketers, students, researchers — anyone who reads, thinks, and writes for a living and wants the AI to remember what they've learned.

What makes it different: most AI tools forget everything when you close the tab. This one compiles what you feed it into a persistent wiki and carries context from one day to the next.

👉 New here? Jump to First 15 minutes below.


Table of contents


What is a second brain?

A "second brain" is a place where you store everything you want to remember — notes, ideas, articles you read, decisions you made — so you don't have to hold it all in your head. The concept was popularized by Tiago Forte in his 2022 book Building a Second Brain.

Traditional second brains (Notion, Obsidian, Evernote) are storage systems. You put stuff in, you look stuff up. The AI-powered version adds something new: the tool actually reads, organizes, and connects your material for you — and remembers across sessions.

What this does, in plain English

Imagine you:

  • Read an article this morning and want to remember the key ideas.
  • Made a decision yesterday and want to revisit it next month.
  • Dropped a thought in the shower and want it captured before it's gone.
  • Are starting a project tomorrow and want AI to remember the context next week.

With second-brain-starter, you type simple commands in Claude Code:

  • /ingest <article-url> → Claude reads it, summarizes, saves it.
  • /braindump "random thought about X" → Claude categorizes and connects it.
  • /wiki-build → Claude compiles everything you've fed it into a searchable wiki.
  • /daily-briefing → Claude shows you where you left off yesterday.

Every skill writes to plain markdown files on your disk. Nothing is locked in a cloud you can't export from.

How it works

flowchart LR
    R["Raw sources<br/>URLs · PDFs · text · videos"]
    S["_sources/<br/>your inbox"]
    W["_wiki/<br/>cross-linked pages<br/>index · log · contradictions"]
    F["Skills<br/>/focus · /daily-briefing<br/>/content-idea · /weekly-review"]
    A["Hooks + cron<br/>session-end · pre-compact<br/>heartbeat · lint"]

    R -->|/ingest| S
    S -->|/wiki-build| W
    W --> F
    A -.keeps state alive.-> W
    A -.keeps state alive.-> F

Prerequisites

You need three things before installing:

1. Claude Code (the CLI tool from Anthropic)

Claude Code is a command-line tool that lets you chat with Claude and have it edit your files.

Install it:

npm install -g @anthropic-ai/claude-code

Don't have Node.js / npm? Install from nodejs.org first (pick the LTS version).

Verify it worked:

claude --version

You'll also need an Anthropic account with API access or a Claude Pro/Max subscription — follow the setup at docs.anthropic.com/claude-code.

2. jq (a tiny JSON processor used by the installer)

OSCommand
Ubuntu / Debian / WSLsudo apt install jq
Fedora / RHELsudo dnf install jq
macOSbrew install jq
Windows (WSL)same as Ubuntu

3. crontab (comes pre-installed on Linux/macOS)

Check it's there:

crontab -l

If you see "no crontab for...", that's fine — it just means you don't have any scheduled jobs yet. install.sh will add two.

On WSL, cron isn't running by default. Start it with:

sudo service cron start

Not using Linux or macOS?

  • Windows: use WSL2. Native Windows is not supported.
  • ChromeOS: works inside the Linux container.

Install

Three commands total:

# 1. Download the repo to somewhere you can write to (no sudo needed)
git clone https://github.com/marciohideaki/second-brain-starter.git ~/second-brain

# 2. Go into the folder
cd ~/second-brain

# 3. Run the installer
./install.sh

The installer is idempotent — safe to run as many times as you want. It will never overwrite your personal data.

What the installer does (all reversible):

  1. Creates shortcuts to 14 skills so Claude Code recognizes them in any folder.
  2. Adds 4 small scripts that run automatically when Claude starts/ends a session.
  3. Adds 2 scheduled tasks to your crontab (a daily health check and a weekly lint).
  4. Appends a "Second Brain" section to ~/.claude/CLAUDE.md so Claude reads your vault at session start.

See Troubleshooting below if anything goes wrong.


First 15 minutes

After install.sh finishes, open any terminal and type:

claude

This opens a Claude Code session. Now try this flow:

Minute 1-3 — Personalize your vault

Type in the Claude session:

/init

Claude will ask 5 quick questions (name, field, goal, tools). It fills in your _knowledge/about-me.md and _knowledge/goals.md automatically.

Minute 3-7 — Capture a thought

Got something on your mind? Type:

/braindump I've been thinking about launching a newsletter. Not sure if weekly or biweekly. Topic would be my experiments with AI tools.

Claude organizes it, flags it as an idea, and connects it to your goals.

Minute 7-12 — Ingest your first source

Paste an article URL, video transcript, or raw text:

/ingest https://example.com/an-article-i-just-read

Claude reads it, shows you a summary + key takeaways, and asks if you want to save it. You approve → it lands in _sources/.

Repeat with 2-3 more sources over the next days.

Minute 12-15 — Compile your wiki

Once you have a few sources:

/wiki-build

Claude extracts concepts and entities from your sources, proposes wiki pages with cross-references, shows you the plan, and asks for approval. Your _wiki/ folder fills up with cross-linked knowledge.

Day 2 and beyond

Every morning:

/daily-briefing

Every time you end work:

/end-session

Every Monday:

/weekly-review

That's the rhythm. Everything else (hooks, cron jobs) happens in the background without you thinking about it.


The 14 skills

SkillWhat it doesExample
/initPersonalizes your vault in 5 questions/init
/braindump [text]Captures a loose thought, categorizes, connects to prior notes/braindump I keep bouncing between ideas
/ingest [source]Reads a URL/file/text, summarizes, saves after your approval/ingest https://blog.com/article
/wiki-buildCompiles all your sources into cross-linked wiki pages/wiki-build
/focus [project]Loads minimal context for one specific project/focus my-newsletter
/daily-briefingMorning summary: priorities, active projects, recent decisions/daily-briefing
/weekly-reviewWeekly reflection: what moved, what stalled, next week's top 3/weekly-review
/end-sessionCloses session, saves learnings, updates project state/end-session
/session-handoffWrites HANDOFF.md so tomorrow's session resumes cleanly/session-handoff
/content-idea [topic]Generates 3-7 content ideas grounded in YOUR real material/content-idea LinkedIn post
/justify [proposal]Checks a proposal against prior decisions, learnings, and wiki pages/justify "Adopt tool X"
/predict [project]Predicts likely next tasks from project state, roadmap, and work log/predict my-project
/lintChecks vault health, flags broken links and stale notes/lint
/skill-improve [path]Runs an AI self-improvement loop on any skill/skill-improve _bootstrap/global/commands/content-idea.md

See docs/en/skills-reference.md for every skill with sample outputs.


How this differs from other tools

Obsidian (plain)Notion AIChatGPT / NotebookLMNicholasSpisak/second-brainsecond-brain-starter
Auto-organizes your notes❌⚠️❌✅✅
Remembers between sessions✅ (manual)⚠️❌✅✅
Warns you when you skipped a step❌❌❌❌✅
Checks vault health automatically❌❌❌❌✅
Improves its own skills over time❌❌❌❌✅
Runs 100% locally on your files✅❌❌✅✅
One-command install———npx skills add./install.sh

Multi-agent support

Primary support is Claude Code (the richest set of hooks + native slash-commands). Adapters exist for Cursor, Gemini CLI, Codex CLI, and Antigravity at different maturity levels.

AgentStatusInstall
Claude Codestable (default)./install.sh
Cursorfunctional./install.sh --agent=cursor
Gemini CLIbeta (SSOT + cron only)./install.sh --agent=gemini-cli
<<<<<<< Updated upstream
Codex CLIfunctional (skills + AGENTS.md + cron, no hooks)./install.sh --agent=codex
Antigravitylightweight (AGENTS.md + cron)./install.sh --agent=antigravity
=======
Codex CLIfunctional (AGENTS.md + global skills + cron)./install.sh --agent=codex
Antigravitystub (community PR welcome)./install.sh --agent=antigravity

Stashed changes

What you lose outside Claude Code: the 4 event hooks (auto-detection of skipped /end-session, pre-compaction state snapshots, prompt logging, long-op notifications). Cron jobs and skill definitions still work in every adapter.

Full details + how to contribute a stronger adapter: docs/en/multi-agent.md.


Philosophy

Three ideas drive the design:

  1. Compile knowledge once, at ingestion. Don't make the AI re-read the same documents every time you ask a question. Pattern borrowed from Karpathy's LLM Wiki.

  2. Session continuity is infrastructure, not a skill. Hooks and cron jobs keep state alive between sessions — you don't have to remember to run anything.

  3. Skills should measurably improve themselves. The /skill-improve loop runs systematic A/B tests on any skill and keeps only the mutations that score better. Based on the autoresearch loop pattern.

Read more: docs/en/philosophy.md.


FAQ

"Do I need to be a developer to use this?"

No. You need to be comfortable copy-pasting commands into a terminal for the first 15 minutes (the install + first session). After that, you only type slash-commands in a Claude chat, which is no different from using ChatGPT.

"Do I need to pay for Claude?"

Yes. Claude Code needs an Anthropic account — either pay-as-you-go API credits, or a Claude Pro/Max subscription. See claude.com/pricing.

"Does this work on Windows?"

Only through WSL2 (Windows Subsystem for Linux). Native Windows is not supported. WSL is free and takes about 10 minutes to install.

"Where does my data live?"

On your disk, in the folder where you cloned this repo. Every file is plain markdown. You can open the folder in Obsidian, Logseq, VS Code, or any text editor. Nothing is in the cloud.

"Is my content sent to Anthropic?"

When you run a skill, Claude Code sends the relevant files to Anthropic's API to process — same as any other AI chat. When no skill is running, nothing is sent.

"Can I use this without any of the cron jobs or hooks?"

Yes. They are all optional. See docs/en/hooks-and-crons.md for how to disable pieces.

"Can I use my own GitHub repo to version this?"

Yes. Fork or clone, and push to your own remote. Your personal notes in _sources/, _wiki/, _learnings/ get committed with the rest.

"What's the difference between this and just using Claude directly?"

Claude forgets everything between sessions. This starter teaches Claude to read your vault at every session start, so context persists. It also gives you 12 pre-built skills that save you from reinventing prompts.

"Will this work with other AI tools like Cursor, Gemini CLI, etc.?"

<<<<<<< Updated upstream Yes, with different maturity levels. Claude Code is the primary target (all 4 hooks supported). Cursor and Codex CLI have functional adapters (skills/rules + cron, no hooks). Gemini CLI is beta and Antigravity is a lightweight AGENTS.md adapter. Full matrix: docs/en/multi-agent.md.

Yes, with different maturity levels. Claude Code is the primary target (all 4 hooks supported). Cursor and Codex CLI have functional adapters; Gemini CLI is beta; Antigravity is still a stub. Full matrix: docs/en/multi-agent.md.

Stashed changes


Troubleshooting

"Permission denied" when running install.sh

chmod +x install.sh
./install.sh

"jq: command not found"

Install it — see Prerequisites.

"crontab: command not found" or cron not running

  • Linux/macOS: crontab is pre-installed. Run crontab -l once to initialize.
  • WSL: sudo service cron start to start the cron daemon, then re-run ./install.sh.

"I ran install.sh but skills don't appear in Claude Code"

Restart your Claude Code session. Skills are loaded from ~/.claude/commands/ at session start.

"Hooks are not firing"

Check ~/.claude/settings.json contains the hook entries and that hook scripts are executable:

chmod +x _bootstrap/global/hooks/*.sh

"I want to uninstall everything"

# remove skill shortcuts
rm ~/.claude/commands/{braindump,ingest,focus,daily-briefing,weekly-review,end-session,session-handoff,content-idea,lint,init,wiki-build,skill-improve}.md

# remove cron jobs
crontab -e  # delete the heartbeat and lint lines

# (optional) remove the "## Second Brain" block from ~/.claude/CLAUDE.md
# (optional) revert ~/.claude/settings.json from its .bak file

More in docs/en/troubleshooting.md.


Documentation


Contributing

PRs, issues, and feedback welcome. See CONTRIBUTING.md.

Every shell script passes shellcheck via CI. Please keep it that way.

License

MIT — free to use, modify, fork, ship. Attribution appreciated but not required.


🇧🇷 Português (BR)

O que é

Kit inicial para construir um "segundo cérebro" — um sistema de notas que a IA organiza e lembra por você — usando Claude Code.

Diferente do Notion ou Obsidian (que só armazenam), este starter faz a IA ler, organizar e conectar o que você guarda. Diferente do ChatGPT (que esquece tudo a cada sessão), este lembra de um dia para o outro porque escreve tudo em arquivos markdown no seu disco.

Para quem: desenvolvedores, criadores de conteúdo, marketing, publicitários, influencers, estudantes, pesquisadores — qualquer pessoa que lê, pensa e escreve todos os dias.

O que você consegue fazer

  • /ingest <url-de-artigo> → Claude lê, resume, salva.
  • /braindump "pensamento solto" → Claude categoriza e conecta.
  • /wiki-build → Claude compila tudo num wiki cruzado.
  • /daily-briefing → Claude mostra onde você parou ontem.
  • /content-idea LinkedIn → Claude gera ideias baseadas no SEU material real.

Tudo em markdown no seu disco. Você pode abrir com Obsidian, VS Code, ou qualquer editor.

Pré-requisitos

  1. Claude Code (CLI da Anthropic): npm install -g @anthropic-ai/claude-code (precisa Node.js)
  2. jq (processador de JSON): sudo apt install jq (Ubuntu/WSL) / brew install jq (macOS)
  3. crontab (já vem no Linux/macOS; no WSL: sudo service cron start)
  4. Conta Anthropic com créditos de API ou assinatura Claude Pro/Max — veja claude.com/pricing.

Windows: só funciona via WSL2. Windows nativo não tem suporte.

Instalação

Três comandos:

git clone https://github.com/marciohideaki/second-brain-starter.git ~/second-brain
cd ~/second-brain
./install.sh

Depois, em qualquer sessão do Claude Code:

/init

Primeiros 15 minutos

  1. /init — wizard de 5 perguntas, personaliza o vault.
  2. /braindump "ideia qualquer" — captura e organiza um pensamento.
  3. /ingest https://algum-artigo.com — ingere uma fonte externa com sua aprovação.
  4. Repita /ingest com mais 2-3 fontes ao longo dos dias.
  5. /wiki-build — compila todas as fontes num wiki cruzado.

Depois: /daily-briefing no começo do dia, /end-session ao terminar, /weekly-review nas segundas.

Skills (12)

/init, /braindump, /ingest, /wiki-build, /focus, /daily-briefing, /weekly-review, /end-session, /session-handoff, /content-idea, /lint, /skill-improve.

Referência completa: docs/pt-br/skills-reference.md.

Filosofia em 3 frases

  1. Compila o conhecimento uma vez, na ingestão — depois responde do wiki (padrão Karpathy).
  2. Continuidade de sessão é infraestrutura, não skill — hooks e cron cuidam disso.
  3. Skills devem se melhorar de forma mensurável — /skill-improve roda o autoresearch loop.

Suporte multi-agente

Suporte primário é Claude Code (4 hooks + slash-commands nativos). Existem adapters para Cursor, Gemini CLI, Codex CLI e Antigravity em níveis diferentes de maturidade.

AgenteStatusInstalação
Claude Codeestável (default)./install.sh
Cursorfuncional./install.sh --agent=cursor
Gemini CLIbeta (SSOT + cron)./install.sh --agent=gemini-cli
<<<<<<< Updated upstream
Codex CLIfuncional (skills + AGENTS.md + cron, sem hooks)./install.sh --agent=codex
Antigravityleve (AGENTS.md + cron)./install.sh --agent=antigravity
=======
Codex CLIfuncional (AGENTS.md + skills globais + cron)./install.sh --agent=codex
Antigravitystub (PR welcome)./install.sh --agent=antigravity

Stashed changes

Fora do Claude Code você perde os 4 hooks de evento. Crons e skills continuam funcionando.

Detalhes: docs/pt-br/multi-agent.md.

FAQ (resumida)

Preciso ser programador? Não, só precisa copiar-colar 3 comandos no terminal na instalação. Depois é só chat.

Preciso pagar pelo Claude? Sim, conta Anthropic com créditos ou assinatura Claude Pro/Max.

Funciona no Windows? Só via WSL2. Windows nativo não.

Meus dados vão para algum lugar? Ficam no seu disco, em markdown. Quando você roda uma skill, o Claude Code envia os arquivos relevantes para a API da Anthropic processar (igual qualquer chat com Claude).

Posso desligar os hooks/crons? Sim, todos são opcionais. Ver docs/pt-br/hooks-and-crons.md.

FAQ completa: docs/pt-br/faq.md.

Documentação

Licença

MIT — livre para usar, modificar, forkar, publicar. Atribuição bem-vinda mas não obrigatória.

Ver o original no GitHub

Bom para

Relacionados