convez.ai
SkillAutomaçãoIntermediário
webdevtodayjason

Dev-Kit — Workflow Portável para Claude Code

webdevtodayjason/dev-kit

Workflow portável e seguro para desenvolvimento com Claude Code

Ver no GitHub
InicialPluginScore · de 0 a 100

Plugin para Claude Code que configura um fluxo de desenvolvimento completo com skills, comandos, sub-agentes, hooks de segurança e template CLAUDE.md. Instala-se uma única vez, não requer credenciais hardcoded e permite total controle sobre permissões e integração com ferramentas próprias.

Instalação

$npx convez-ai webdevtodayjason/dev-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

  • Onboarding de agentes de código com contexto e segurança
  • Configuração de hooks de segurança e validação de comandos destrutivos
  • Integração com MCP (context7, Linear, ghost.build) sem expor credenciais

Veredito

Skill sólida para quem quer padronizar workflows em Claude Code com segurança e portabilidade. A abordagem zero-credentials e a cobertura de hooks destrutivos a destacam, mas exige leitura cuidadosa da documentação e setup inicial com vários pré-requisitos.

README

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

dev-kit — a portable Claude Code workflow

A one-install Claude Code marketplace plugin that sets up a complete development workflow: skills, slash commands, sub-agents, output styles, optional security hooks, and a fill-in CLAUDE.md template.

Ships zero credentials. No API keys, no private infrastructure, no machine-specific paths. You bring your own MCP keys and choose your own permission posture.


⚡ Quick setup — paste this to your agent

Hand this to your Claude Code session (or any coding / cloud agent) and it will set everything up, interviewing you only where it needs your specifics:

You're setting up the "dev-kit" Claude Code marketplace plugin on my machine.
Work through these steps in order, asking me only when you genuinely need input:

1. INSTALL: run `/plugin marketplace add webdevtodayjason/dev-kit`, then
   `/plugin install dev-kit@jason`, then tell me to restart Claude Code so the
   skills, commands, agents, and the context7 MCP register.

2. PREREQS (report what's missing; don't install anything without asking first):
   - node / npx  -> needed for the keyless context7 docs MCP
   - uv          -> needed only for the optional damage-control security hooks
   - gh CLI (authenticated) -> needed for the review/commit workflows and find-docs

3. CLAUDE.md: copy `plugins/dev-kit/CLAUDE.template.md` to `~/.claude/CLAUDE.md`
   (or this project's `./CLAUDE.md`). Then interview me briefly to fill every
   <PLACEHOLDER> (my name, machine/resources, where my secrets live, current
   projects, my follow-up-list file path). Delete any section whose tooling I
   have not installed.

4. CONSERVATIVE PERMISSIONS: in `~/.claude/settings.local.json`, add a deny list
   for catastrophic commands (rm -rf /*, rm -rf ~/*, sudo rm -rf, mkfs,
   dd ... of=/dev/*) and an ask list (git push --force, git reset --hard).
   Do NOT enable bypassPermissions, CLAUDE_AUTONOMY, or any skip-prompt flags.

5. MCP CREDENTIALS (mine only — never reuse anyone else's): context7 works
   keyless out of the box. If I use Linear or ghost.build, walk me through adding
   them with MY own key / login.

6. OPTIONAL SECURITY HOOKS: if I want the damage-control PreToolUse firewall,
   confirm uv is installed, then run the damage-control skill's install cookbook.

7. VERIFY: run `/context-audit` and confirm the new skills and commands appear.

Never write secrets into a committed file. Confirm each step's result before
moving to the next.

Install (manual)

/plugin marketplace add webdevtodayjason/dev-kit
/plugin install dev-kit@jason

Then restart Claude Code so skills, commands, agents, output styles, and the context7 MCP register.

Prerequisites (install these yourself first)

ToolNeeded forInstall
Node / npxthe keyless context7 docs MCPcomes with Node.js
uvthe optional damage-control security hooks (Python)brew install uv
gh CLI (authed)the review/commit workflows + find-docsbrew install gh && gh auth login
tmuxonly if you later add team/orchestrator panesbrew install tmux

None are required just to install — the skills and commands work without them; the table notes what each specific feature needs.


After install — 3 things to do

1. Adopt the CLAUDE.md template

Open plugins/dev-kit/CLAUDE.template.md (from this repo, or from the installed plugin directory under ~/.claude/plugins/) and copy it to your global or project config:

cp plugins/dev-kit/CLAUDE.template.md ~/.claude/CLAUDE.md   # or a project ./CLAUDE.md

Then replace every <PLACEHOLDER> (your name, environment, secrets location, follow-up file path, etc.) and delete any section whose tooling you haven't installed.

2. Choose a permission posture (start conservative)

This kit deliberately ships no settings.json. Set your own — and do not copy a "full bypass" setup. Recommended starting point in ~/.claude/settings.local.json:

{
  "permissions": {
    "deny": [
      "Bash(rm -rf /*)", "Bash(rm -rf ~/*)", "Bash(sudo rm -rf:*)",
      "Bash(mkfs:*)", "Bash(dd if=* of=/dev/*)"
    ],
    "ask": [ "Bash(git push --force:*)", "Bash(git reset --hard:*)" ]
  }
}

Leave defaultMode at the prompting default (or at most acceptEdits). Do not set CLAUDE_AUTONOMY, skipDangerousModePermissionPrompt, or skipAutoPermissionPrompt until you've deliberately decided you want them — they disable safety prompts and neuter the deny/ask rules above.

3. Bring your own MCP credentials

  • context7 (docs lookup) — ships keyless and works out of the box. Optionally add your own CONTEXT7_API_KEY for higher rate limits.
  • Linear — add the public server with your own key/OAuth (not bundled).
  • ghost.build — install the current ghost CLI yourself and run ghost login for your own account (not bundled).
  • llm-council skill — set OPENAI_API_KEY / GEMINI_API_KEY in your own .env (the shipped .env.template has only placeholders).

Optional: enable the damage-control security hooks

The damage-control skill ships with a self-contained PreToolUse firewall (blocks destructive shell commands and writes to credential paths). It's not auto-wired so it can't break your tool calls on day one. To turn it on, install uv, then run the damage-control skill and follow its install cookbook.


What's included

  • Skills (24): tdd, diagnose, find-docs, write-a-skill, to-issues, to-prd, triage, code-review-style review workflows, grill-me, grill-with-docs, improve-codebase-architecture, context-audit, caveman, zoom-out, setup-matt-pocock-skills, damage-control, followup-closet, obsidian-vault, ghost-db, llm-council, prod-db-surgical, spec-driven-dev, and node / python / rust conventions.
  • Commands: /code-review, /ai-slop-cleaner, /security-review, /commit, /geo-optimize.
  • Agents: bug-analyzer, code-reviewer, dev-planner, story-generator, ui-sketcher.
  • Output styles: coding-vibes, structural-thinking.
  • MCP: context7 (keyless docs lookup).
  • CLAUDE.template.md: fill-in operating instructions.
  • Statusline: plugins/dev-kit/statusline/statusline.sh — a clean Claude Code status bar (project · branch · model · context % · tokens).
  • Terminal preset: plugins/dev-kit/terminal/starship.toml + TERMINAL-SETUP.md — iTerm2 + Nerd Font + Starship, with an AI handoff prompt.

What's intentionally NOT included

Anything machine-specific or private: orchestration tooling, secret managers, personal infrastructure, business-project specifics, and any credentials. Bring your own keys and your own conventions where they differ.


Companion guides

  • ONBOARDING.md — full macOS new-dev walkthrough (install Claude → paste-to-agent setup of the whole toolchain).
  • TERMINAL-SETUP.md — iTerm2 + Nerd Font + Starship + the Claude statusline, with an AI handoff prompt.
  • STACK.md — the companion plugins (claude-mem, CodeRabbit, Codex, Ralph Wiggum) + MCPs (Linear, ghost) that complete the workflow, with exact install commands.

Built by Jason Brashear — jasonbrashear.com · Substack · GitHub @webdevtodayjason

Ver o original no GitHub

Bom para

Relacionados