Leia em Português
Mind — template
A personal knowledge base organized as a tree of Markdown files, meant to be edited visually in an Obsidian-like editor and queried on demand by Claude Code — Anthropic's official CLI. This repository only really works alongside it: it's claude running in your terminal that reads the Skill and the capture trigger here, and decides when to look something up or save it. Without Claude Code installed, the vault is just a plain folder of Markdown.
This repository is the raw/template version: only the engineering (Skill, capture trigger, symlink script, architecture docs), with no personal content. Clone it, personalize it, and start filling it with your own nodes.
The full architecture (why each decision was made, what's still missing) is in docs/ARQUITETURA.md. This README is just the "how to use it".
Prerequisites
- Claude Code installed and configured (
claude available in your terminal) — see the official install guide at the link.
- Optional: a Markdown editor with graph/backlink support, like Obsidian, to navigate and edit nodes visually. Not required — you can use it from the terminal alone.
What's in here
MIND.md: root index, starts empty. There's no fixed pre-built folder tree — nodes (and whatever folders make sense to organize them) get created gradually, as conversations happen.persona.md (created on demand, never shipped): an optional "mirror persona" — a profile of how you talk and act, built gradually by Claude from your real conversations so it answers in your own style. Opt-in, with levels (off / voice / voice+behavior / +humor), chosen in config.md. It's personal content: it lives in your private repo. See docs/ARQUITETURA.md, section 14..claude/: Skill and permissions that only apply while this project is active.claude-user/: the "warehouse" for user-level Skill, Subagents, and instructions — they work in any active project, not just this one. They live here (versioned in this repo) and get mirrored into ~/.claude/ via symlink (see below).scripts/setup-symlinks.sh: recreates the claude-user/ symlinks in ~/.claude/.scripts/status-all.sh: scans whatever project repos you list inside it and shows which ones have uncommitted changes or unpushed commits — to check everything at once instead of going folder by folder.scripts/check-template-version.sh: reports whether your engineering base is behind this template (read-only).VERSION / CHANGELOG.md: version number of the engineering base and what changed in each bump — used by the check above and by the maintenance round.
Setup on a new machine
Clone it as a "manual" fork via git (two remotes) — that way you can pull future updates
from the template (section below) without losing your own nodes. Don't use GitHub's
"Use this template" button: it creates a git history disconnected from the original and
breaks this flow (full reasoning in docs/ARQUITETURA.md, section 12).
-
Create an empty, private repository (GitHub, GitLab, etc.) to hold your vault —
don't initialize it with a README/license/gitignore, it needs to be truly empty.
-
Clone the template and re-point it to that empty repo — use SSH if you already have a
key configured on GitHub (won't ask for a password afterward), or HTTPS if you don't
(always works, but asks for username/token on every push). You can mix protocols:
upstream on one and origin on the other, no problem.
Via SSH:
git clone [email protected]:CafeLabsCorp/mind-template.git mind
cd mind
git remote rename origin upstream
git remote add origin <ssh-url-of-your-empty-repo>
git push -u origin main
Via HTTPS:
git clone https://github.com/CafeLabsCorp/mind-template.git mind
cd mind
git remote rename origin upstream
git remote add origin <https-url-of-your-empty-repo>
git push -u origin main
-
Run the setup and start using it:
./scripts/setup-symlinks.sh
claude
You can clone it under any name and at any path — there's no fixed-location requirement, unless you're using a voice integration (see section below).
Running ./scripts/setup-symlinks.sh again is safe at any time (idempotent) — only needed when a new Skill/Subagent gets added under claude-user/.
From there it's just conversation: ask it to look something up, or mention something worth keeping — the Skill and the trigger handle the rest (see section below).
Receiving template updates
This repository evolves (new Skill rules, architecture tweaks). Since the setup above already configured upstream, pulling updates later is always:
git fetch upstream
git merge upstream/main
This usually merges without conflict because the architecture already separates "engineering" (docs/, claude-user/, scripts/, .claude/) — which only changes here, in the template — from personal content (your filled-in MIND.md and whatever node folders you create) — which the template never touches. The expected friction point is .claude/settings.json: it accumulates permissions granted as you use it, so if you and the template change that file at the same time you can get a merge conflict there — rare, and resolves by hand without mystery (small file). More details on this decision in docs/ARQUITETURA.md, section 12.
以上为 README 的开头部分。其余内容请阅读原文。 查看完整 README ↗
