dotclaude – Setup Portátil para Claude Code
Okohedeki/dotclaude
Leve seu ambiente Claude Code completo para qualquer máquina em um comando
Ver no GitHubRepositório que centraliza configurações, skills customizadas e claude-mem em um único setup reproduzível. Com um comando em máquina nova, você recupera todo seu ambiente Claude Code sem perder contexto local. Usa symlinks para manter settings sincronizadas com git, enquanto mantém memória e sessões locais.
Instalação
npx convez-ai Okohedeki/dotclaude1 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
- Sincronizar configurações e skills do Claude Code entre múltiplas máquinas
- Onboarding rápido de ambiente de desenvolvimento com Claude integrado
- Compartilhar skills customizadas via git mantendo memória pessoal local
Veredito

Solução madura e bem pensada para portabilidade de ambiente Claude Code. Implementação idempotente com boas práticas (symlinks para settings, preservação de backups). Ideal para desenvolvedores que usam múltiplas máquinas ou querem versionamento de skills.
README
A tradução para o português deste README ainda está na fila.
# fresh machine, one command (clones + runs setup.sh):
curl -fsSL https://raw.githubusercontent.com/Okohedeki/dotclaude/main/bootstrap.sh | sh
# …or do it by hand:
git clone https://github.com/Okohedeki/dotclaude.git ~/Documents/dotclaude
cd ~/Documents/dotclaude && ./setup.sh
# → settings + skills symlinked into ~/.claude/
# → claude-mem installed via npx
# → memory/ left alone (machine-local, on purpose)
Sync the parts of ~/.claude/ that travel — settings, hand-authored skills, the plugins manifest — while leaving the parts that don't (memory, sessions, caches, personal in-progress plans) firmly local. No vendored dependencies; no fork of anyone's plugins. Just the small handful of files you actually edit, plus an idempotent installer that bootstraps a new machine.
- Symlinked, not copied: edits in
~/.claude/settings.jsonflow back to the repo automatically.git statusshows your drift; commit it when you're happy. - Skill authoring is live: two hand-authored seed skills (
rebrand-project,refresh-readme-from-reference), seven curated fromalirezarezvani/claude-skills, and four fromboraoztunc/skills(ogilvy,stop-slop,frontend-design,vercel-react-best-practices) live inskills/. New skill = new folder underskills/<name>/SKILL.md, picked up by Claude Code on next session. - claude-mem integrated: the installer runs
npx claude-mem installso persistent memory + the 5 lifecycle hooks are wired up automatically. Upstream maintains the version; bumping is just re-running setup. - Memory stays local:
~/.claude/memory/is never read, copied, or symlinked. Personal context doesn't cross machines. - Safe to re-run: the installer is idempotent. Pre-existing files in
~/.claude/are preserved at<name>.pre-dotclaude.bakon first run so nothing gets clobbered.
Getting Started
git clone https://github.com/Okohedeki/dotclaude.git ~/Documents/dotclaude
cd ~/Documents/dotclaude
./setup.sh
The installer:
- Symlinks
settings/settings.jsonand everyskills/<name>/into~/.claude/. - Copies (does not symlink)
settings/settings.local.jsonand theplugins/*.jsonmanifests only if they don't already exist — these are templates because Claude Code mutates them with machine-specific paths, and a symlink would leak those paths back into the repo. - Runs
npx -y claude-mem install. - Backs up any pre-existing
~/.claude/<name>it would overwrite, to<name>.pre-dotclaude.bak.
Re-running is safe — every step is idempotent.
What's in the repo
dotclaude/
├── README.md this file
├── LICENSE Apache-2.0
├── setup.sh the installer
├── .gitignore keeps personal artifacts out of git
│
├── settings/
│ ├── settings.json portable Claude Code settings (symlinked)
│ └── settings.local.json template — machine-specific (copy-on-install)
│
├── skills/
│ ├── rebrand-project/SKILL.md rename a project end-to-end
│ ├── refresh-readme-from-reference/SKILL.md rewrite a README in another repo's aesthetic
│ ├── chaos-engineering/SKILL.md curated — design + run resilience experiments
│ ├── security-guidance/SKILL.md curated — PreToolUse security-pattern guidance
│ ├── slo-architect/SKILL.md curated — design SLOs + error budgets
│ ├── senior-architect/SKILL.md curated — system design + architecture review
│ ├── senior-security/SKILL.md curated — threat modeling + security audit
│ ├── senior-secops/SKILL.md curated — vuln assessment + compliance
│ └── grill-with-docs/SKILL.md curated (MIT) — docs-anchored plan grilling
│
├── plugins/
│ ├── installed_plugins.json template — which plugins to enable
│ └── known_marketplaces.json template — marketplaces to subscribe to
│
├── plans/ gitignored except for .gitkeep
└── assets/logo.svg repo brand mark
Skills
Thirteen skills ship in the repo today: two hand-authored seeds, seven curated from the alirezarezvani/claude-skills library (Apache-2.0 except grill-with-docs, which is MIT), and four from boraoztunc/skills — ogilvy, stop-slop, frontend-design, vercel-react-best-practices. Each third-party skill keeps its own terms; see its SKILL.md / LICENSE (e.g. frontend-design ships a LICENSE.txt; stop-slop and vercel-react-best-practices are MIT).
Hand-authored
rebrand-project
Rename a project across its codebase, JSON schema, file extension, well-known paths, HTTP headers, A2A/extension URIs, GitHub repo, npm package, CLI, and live deployment, in one breaking-version cut. Covers the full workflow from inventory pass → naming-table lock-in → mechanical edits → migration doc/ADR → local verification → publication. Distilled from a real production rename (Okohedeki/airlock → Okohedeki/airlock-config, ~425 references, 127 tests, zero regressions).
refresh-readme-from-reference
Rewrite a project's README in the visual aesthetic of a reference repo (paseo, anthropic, supabase, etc.) — centered HTML hero, badge row, hero image or code block, selling-point bullets, paseo-style section order — while keeping the project's actual feature claims and code examples intact. Uses curl -sL, not WebFetch, because WebFetch strips the HTML structure you need to study.
Authoring new skills
mkdir -p skills/<new-skill-name>
$EDITOR skills/<new-skill-name>/SKILL.md
Format mirrors the official skill-creator template — YAML frontmatter (name, description) + markdown body. Once you save, the symlinked ~/.claude/skills/<new-skill-name>/ is live — Claude Code picks it up on next session.
If you want help authoring, the official Anthropic skill-creator plugin walks you through it (and runs evals).
claude-mem
thedotmack/claude-mem — persistent memory compression across Claude Code sessions, plus five lifecycle hooks (SessionStart, UserPromptSubmit, PostToolUse, Stop, SessionEnd). It's not vendored here; the installer just runs npx -y claude-mem install, so version bumps stay with upstream. Its own settings live at ~/.claude-mem/settings.json — see the upstream README for configuration options.
Memory is off-limits
~/.claude/memory/ is never read, copied, or symlinked by this repo, not now and not in any future change. Personal/work memory stays local to each machine where it was written. If you want memory on a second machine, write it there fresh — the design is intentional, not an oversight.
Development
# After cloning, before pushing changes to settings/skills:
./setup.sh # link the repo into ~/.claude
$EDITOR skills/<name>/SKILL.md # author or edit a skill
git status # see drift between repo and live config
git commit -am "Add <name> skill"
git push
To verify the setup actually wired up correctly:
ls -la ~/.claude/skills/ # should show symlinks into ~/Documents/dotclaude/
ls -la ~/.claude/settings.json # should be a symlink into the repo
ls ~/.claude-mem/ # claude-mem's own dir, created by its installer
License
Apache-2.0 — same as the upstream Anthropic plugin marketplace and claude-mem. Fork freely.