convez.ai
SkillProdutividadeIntermediário
inf0rmer

Brain Kit

inf0rmer/brain-kit

Framework local para segundo cérebro em Obsidian via Claude Code

Ver no GitHub
InicialPluginScore · de 0 a 100

Meta-framework que scaffolda vaults locais de Obsidian como "segundo cérebro" acionados por Claude Code skills. Permite criar brains independentes com CLI intuitivo, captura agentic de informações (sem embeddings) e publicação controlada de notas.

Instalação

$npx convez-ai inf0rmer/brain-kit

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
28%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Construir um segundo cérebro pessoal com notas estruturadas e navegação agentic
  • Capturar, revisar e recuperar informações via conversas com Claude
  • Automatizar sínteses de 1-1s e digests de status a partir de notas

Veredito

Solução elegante e bem-pensada para PKM (personal knowledge management) que aproveita a navegação agentic do Claude em vez de embeddings. Excelente para quem trabalha com Obsidian e quer automatização inteligente mantendo dados locais e privados.

README

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

brain-kit

The meta framework for a personal "second brain" backed by Obsidian and driven by Claude Code. This repo is not a brain — it's the kit you use to init brains, each living in its own separate repository.

What's here

brain-kit/
  bin/brain             # the `brain` CLI dispatcher (put bin/ on your PATH)
  libexec/
    brain-init          #   `brain init` — scaffolds a new brain vault
    brain-open          #   `brain open` — opens a vault in Obsidian
    brain-update        #   `brain update` — re-syncs framework files into a vault
  skills/               # canonical brain skills (copied into each new brain)
    brain-capture/      #   file new input (text or a Google Doc link)
    brain-recall/       #   answer questions from notes, with [[citations]]
    brain-review/       #   1-1 prep & status digests
    brain-import/       #   pull Google Docs via `gws` (incl. latest-dated-entry)
    brain-publish/      #   publish one note to your Google Docs (gated, confirm-first)
    brain-upkeep/       #   hygiene pass + index refresh (cron-friendly)
  template/             # the vault skeleton copied into a new brain
    CLAUDE.md           #   the brain's operating manual (retrieval/capture protocol)
    .obsidian/          #   baseline Obsidian config (Templates + Bases enabled)
    meta/templates/     #   note templates per type
    meta/indexes/       #   MOC dashboards: static lists (Claude) + embedded Base (you)
    meta/bases/         #   native Obsidian Bases (live dashboards over frontmatter)
    _seed/              #   optional example notes (brain init --seed)
  .claude/skills/
    brain-init/         # the only skill active in the kit (wraps `brain init`)

The brain CLI is a dispatcher: each subcommand is a libexec/brain-<name> script, so future vault-management commands are added by dropping a new script in.

Design in one paragraph

Retrieval is agentic, not embeddings-based: Claude navigates a well-structured vault (consistent frontmatter, [[wikilinks]], index notes) and reasons over grep results. The conventions live in each brain's CLAUDE.md, which is the real "brain". Brains are local-only (sensitive management data never leaves the machine). Content leaves only via the explicit, confirm-first brain-publish skill (publish one note to your own Google Docs, sensitivity-gated); gws is otherwise import-only.

Create a brain

brain init ~/Code/my-brain --name "My Brain"        # clean vault
brain init ~/tmp/demo-brain --name "Demo" --seed     # with example notes

Then:

cd ~/Code/my-brain && claude
# "log my 1-1 with Jane: ..."   /   "prep me for my 1-1 with Jane"
gws auth login        # one-time, enables brain-import from Google Docs

The brain command is available once bin/ is on your PATH (run brain or brain help to list subcommands).

Obsidian

Each brain ships a baseline .obsidian/ config, so it opens ready to use:

  • Core Templates enabled, pointed at meta/templates/.
  • Bases (Obsidian's native database, requires Obsidian 1.9+) enabled. The meta/bases/*.base files render live people / projects / ideas / 1-1 dashboards from note frontmatter. They're embedded at the top of each meta/indexes/*.md.
  • New notes default to 00-inbox/; [[wikilinks]] kept as the link format.

Open a vault in Obsidian:

cd ~/Code/my-brain && brain open      # or: brain open ~/Code/my-brain
brain init ~/Code/my-brain --open     # scaffold and open in one step

The first time, register the vault via Obsidian's Open folder as vault (the command prints the path); after that brain open opens it directly.

Dual layer, on purpose: Bases are for you in the GUI. Claude can't render them, so it reads the plain-markdown index lists + frontmatter instead. Both stay in sync because they're driven by the same frontmatter; brain-upkeep maintains the static lists, Bases self-update.

Automating upkeep (per brain)

So brain-upkeep runs without you remembering:

  • Weekly cron — in Claude Code, schedule a job that opens the brain and runs /brain-upkeep (e.g. Monday 09:00). It refreshes the indexes and writes meta/upkeep-report.md for you to review.

  • SessionStart nudge (optional) — add to the brain's .claude/settings.json:

    {
      "hooks": {
        "SessionStart": [
          { "hooks": [ { "type": "command",
            "command": "grep -rl 'status: active' . | head -1 >/dev/null && echo 'brain: run /brain-upkeep to refresh indexes & flag stale notes'" } ] }
        ]
      }
    }
    

Updating brains after changing the kit

The kit is the source of truth; vaults hold copies of the framework files. To pull the latest kit changes into an existing brain:

brain update ~/Code/my-brain        # or: cd ~/Code/my-brain && brain update
brain update ~/Code/my-brain --dry-run   # preview what would change

brain update re-syncs framework only — skills, note templates, bases, the operating manual (CLAUDE.md, your title preserved), and .gitignore. It never touches your content: notes (people/, projects/, ideas/, resources/, 00-inbox/), your index lists (meta/indexes/), or your Obsidian config (.obsidian/ — pass --force-config to overwrite that too). Review with git diff in the vault and commit when happy.

Don't use brain init --force to upgrade — it would overwrite your index notes and .obsidian/ config. Use brain update.

Ver o original no GitHub

Bom para

Relacionados