convez.ai
SkillAutomaçãoAvançado
jchiam

Claude Config — Configuração Pessoal de Skills e Permissões

jchiam/claude-config

Configuração centralizada de skills, permissões e settings para Claude Code

Ver no GitHub
InicialPluginScore · de 0 a 100

Repositório de configuração centralizado para Claude Code que gerencia settings, permissões de ferramentas e skills customizados. Inclui suite APD com 13 skills especializados (code review, API design, diagramas, documentação) e controles de segurança para operações sensíveis como git push --force e npm publish.

Instalação

$npx convez-ai jchiam/claude-config

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 3 semanas
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Gerenciar permissões de ferramentas CLI (git, npm, gh) de forma segura
  • Organizar e versionar skills customizados reutilizáveis
  • Automatizar tarefas de desenvolvimento (code review, documentação, PRs)
  • Compartilhar configuração entre Claude Code, QoderWork e Qwen VSCode

Veredito

Setup robusto e bem pensado para quem trabalha intensamente com Claude Code. A suite APD oferece skills prontos de alta qualidade. Requer conforto com shell scripts, git e variáveis de ambiente, mas a documentação é clara e o ganho em automação é significativo.

README

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

~/.claude

Personal Claude Code configuration — settings, permissions, and custom skills.

This repo is the source of truth for ~/.claude. A symlink points ~/.claude to this directory. Skills are also synced to other AI tools (QoderWork, Qwen VSCode plugin).

Structure

.
├── settings.template.json     # Committed settings baseline (no secrets)
├── setup.sh                   # macOS/Linux/WSL: full first-time setup
├── setup-links.sh             # macOS/Linux/WSL: configure ~/.claude symlink + skill sync
├── setup-settings.sh          # macOS/Linux/WSL: generate settings.json from template
├── statusline-command.sh      # Status line feature (bash + jq + git)
└── skills/
    ├── open-pr/               # /open-pr skill
    ├── update-docs/           # /update-docs skill
    ├── apd-*/                 # APD skill suite (13 skills)
    ├── documentation-writer/  # Shared skill
    ├── refactor/              # Shared skill
    ├── security-requirement-extraction/  # Shared skill
    └── webapp-testing/        # Shared skill

Settings

settings.json is gitignored — it is generated locally and may contain credentials.

settings.template.json is the committed baseline. It includes all non-sensitive configuration:

  • Global tool permissions (allow/deny/ask lists)
  • Model overrides
  • Plugin and marketplace configuration

Denied (never auto-approved):

  • npm publish — prevent accidental registry publishing
  • git push --force / git push -f — prevent remote history destruction
  • curl * | bash — prevent pipe-to-shell execution
  • npx — prevent arbitrary package execution

Allowed (auto-approved):

  • git -C * — run git commands in any directory without cd
  • Read-only git inspection (git log, git diff, git status)
  • npm introspection (npm list, npm show, npm run)
  • GitHub CLI (gh auth, gh issue, gh pr)
  • Web fetching from trusted domains (github.com, npmjs.com, docs.anthropic.com, unpkg.com)
  • Editing files within ~/.claude/skills/

Generating settings.json

Run the setup script (part of setup.sh):

./setup-settings.sh

This will:

  1. Copy settings.template.json as the base
  2. Prompt for ANTHROPIC_AUTH_TOKEN (skippable — key omitted if blank)
  3. Auto-configure statusLine with the correct path for this machine
  4. Optionally run gt tools configure claude-code to inject OTEL/GovTech settings

Skills

Skills are invoked as slash commands inside Claude Code. They are also synced to QoderWork and Qwen VSCode plugin via setup-links.sh.

Core

SkillCommandDescription
open-pr/open-prPushes branch and opens a PR via gh pr create with structured summary
update-docs/update-docsUpdates README and GitHub wiki to reflect recent repo changes

APD Suite

SkillDescription
apd-api-design-principlesREST and GraphQL API design guidance
apd-brainstormingStructured ideation before building features
apd-code-reviewerAI-powered code review
apd-create-prGitLab MR creation via glab
apd-doc-coauthoringCollaborative documentation writing
apd-draw-ioDraw.io diagram creation and editing
apd-lint-and-validateRun validation after every code change
apd-mermaid-expertMermaid diagram creation
apd-openapi-spec-generationOpenAPI 3.1 spec generation and validation
apd-skill-creatorCreate and optimize new skills
apd-systematic-debuggingStructured debugging before proposing fixes
apd-tddTest-driven development workflow

Shared (from ~/.agents/skills/)

SkillDescription
documentation-writerDiátaxis-guided technical documentation
refactorSurgical refactoring without behavior changes
security-requirement-extractionDerive security requirements from threat models
webapp-testingPlaywright-based local web app testing

Setup

macOS / Linux / WSL

git clone <repo-url> ~/claude_config
cd ~/claude_config
./setup.sh

setup.sh runs both steps in order:

  1. setup-links.sh — creates ~/.claude → this repo symlink, syncs skills to ~/.qoderwork/skills/ and ~/.qwen/skills/
  2. setup-settings.sh — generates settings.json from template, prompts for credentials

Windows (PowerShell)

Clone the repo, then run the following steps in PowerShell.

Step 1: Create junction (run as Administrator, or with Developer Mode enabled)

$repo = "$env:USERPROFILE\claude_config"
$claude = "$env:USERPROFILE\.claude"

if (Test-Path $claude) { Remove-Item $claude -Force -Recurse }
New-Item -ItemType Junction -Path $claude -Target $repo

Step 2: Sync skills (repeat after adding new skills)

$skillsDir = "$env:USERPROFILE\claude_config\skills"
$targets = @("$env:USERPROFILE\.qoderwork\skills", "$env:USERPROFILE\.qwen\skills")

foreach ($targetDir in $targets) {
    if (-not (Test-Path $targetDir)) { New-Item -ItemType Directory -Path $targetDir -Force | Out-Null }
    foreach ($skill in (Get-ChildItem $skillsDir -Directory)) {
        $link = Join-Path $targetDir $skill.Name
        if (Test-Path $link) { Remove-Item $link -Force -Recurse }
        New-Item -ItemType Junction -Path $link -Target $skill.FullName -Force | Out-Null
    }
}
Write-Host "Skills synced."

Step 3: Generate settings.json

$repo = "$env:USERPROFILE\claude_config"
$config = Get-Content "$repo\settings.template.json" -Raw | ConvertFrom-Json

# Set auth token (press Enter to skip)
$token = Read-Host "ANTHROPIC_AUTH_TOKEN (Enter to skip)"
if ($token) {
    $config.env | Add-Member -NotePropertyName "ANTHROPIC_AUTH_TOKEN" -NotePropertyValue $token -Force
}

# Optional: run gt tools configure claude-code after writing
$config | ConvertTo-Json -Depth 20 | Set-Content "$repo\settings.json" -Encoding UTF8
Write-Host "settings.json written."

Status Line (optional, macOS/Linux only)

statusline-command.sh provides an agnoster-style status line showing directory, git branch, model, and context usage. Requires bash + jq + git. Auto-configured by setup-settings.sh.

To add manually (e.g. after gt tools configure overwrites settings.json), add to settings.json:

"statusLine": {
  "type": "command",
  "command": "bash \"/path/to/claude_config/statusline-command.sh\""
}

Windows: Claude Code does not support command-type statusLine on Windows.

Adding New Skills

  1. Create the skill in skills/my-new-skill/SKILL.md
  2. Re-sync: ./setup-links.sh (macOS/Linux) or re-run Step 2 above (Windows)
  3. Commit: git add skills/my-new-skill && git commit -m "feat: add my-new-skill"

Plugins

Plugins extend Claude Code with additional capabilities. Plugin state is managed in plugins/.

Installed

PluginMarketplaceDescription
code-simplifierclaude-plugins-officialReviews changed code for reuse, quality, and efficiency
cavemancavemanUltra-compressed communication mode to reduce token usage

Marketplaces

MarketplaceSource
claude-plugins-officialanthropics/claude-plugins-official (GitHub)
cavemanJuliusBrussee/caveman (GitHub)

plugins/installed_plugins.json and plugins/known_marketplaces.json are committed — they define which plugins and marketplaces are registered. Plugin binaries (plugins/cache/) and runtime data (plugins/data/, plugins/marketplaces/) are gitignored.

What's gitignored

Runtime and ephemeral state generated by Claude Code is excluded:

  • settings.json — generated locally, may contain credentials (use setup-settings.sh to regenerate)
  • backups/, file-history/, sessions/, shell-snapshots/ — transient session data
  • cache/, paste-cache/, session-env/ — runtime caches
  • ide/, plans/, projects/, tasks/ — working state
  • telemetry/, history.jsonl — usage logs
  • memory/ — auto-generated conversation memory
  • plugins/cache/, plugins/data/, plugins/marketplaces/ — plugin runtime (manifests are committed)
  • plugins/blocklist.json, plugins/install-counts-cache.json — internal plugin state
  • .credentials.json — auth secrets
Ver o original no GitHub

Bom para

Relacionados