Agent Setup
DefaultPerson/agent-setup
Setup seguro e notificável para Claude Code & Codex com guards e automações
Ver no GitHubConfiguração universal para Claude Code e OpenAI Codex CLI com sistema de segurança, notificações por áudio, linha de status e comandos produtivos. Bloqueia operações perigosas, fornece alertas de voz em cache, notificações nativas do SO e integra skills/slash commands para fluxos de trabalho acelerados.
Instalação
npx convez-ai DefaultPerson/agent-setup1 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
- Proteger execução de comandos perigosos durante automação com Claude
- Receber notificações de áudio quando Claude conclui tarefas longas
- Automatizar workflows Git (commit, push, PR) com slash commands
- Monitorar status do projeto, branch, modelo e limites de rate em tempo real
Veredito

Setup maduro e multiplataforma que eleva a segurança e UX de agentes Claude/Codex com hooks de permissão, notificações inteligentes e comandos encadeados. Essencial para quem automatiza fluxos de desenvolvimento com IA em produção.
README
A tradução para o português deste README ainda está na fila.
agent-setup
Universal setup for Claude Code and OpenAI Codex CLI — security hooks, notifications, status line, and productivity commands.
Features
- Security Guard — blocks dangerous commands (
rm -rf /, credential reads, privileged docker, etc.) - TTS Notifications — cached voice alerts when Claude finishes or needs input
- Desktop Notifications — native OS notifications (Linux, macOS, Windows)
- Status Line — project, branch, model, context usage, rate limits
- Skills + Slash Commands —
commit(«сделай коммит и пуш» — auto-chain),push-and-pr,/ultrathink,/prime,/publish,/release(+/researchon Codex; on Claude Code use the native/deep-researchskill) - Cross-platform — Linux, macOS, Windows
Claude Code
Setup
Paste into Claude Code — or follow the steps manually:
Prerequisites: Node.js 18+, uv (python package manager), ffmpeg or mpv (audio for TTS).
1. git clone https://github.com/DefaultPerson/agent-setup.git && cd agent-setup
2. cp -r .claude/hooks ~/.claude/hooks
3. cp -r .claude/commands ~/.claude/commands && cp -r .claude/skills ~/.claude/skills
4. cp .claude/settings.example.json ~/.claude/settings.json
# Windows: cp .claude/settings.local.json.windows "$env:USERPROFILE/.claude/settings.json"
5. cp CLAUDE.md ~/.claude/CLAUDE.md (optional — author's coding style and rules)
6. Install recommended plugins (see below)
7. Add shell aliases (see below)
8. Ask me for any preferences
9. Verify everything works
10. Delete agent-setup (repo no longer needed after setup)
Recommended Plugins
LSP, context7, frontend-design are available in the default marketplace — install via /plugin → search.
# Browser automation for AI agents
# https://github.com/vercel-labs/agent-browser
/plugin marketplace add vercel-labs/agent-browser
/plugin install agent-browser@agent-browser
/reload-plugins
Shell Aliases
Add to .bashrc / .zshrc:
alias cc="claude" ccr="claude --resume" ccd="claude --dangerously-skip-permissions" ccdr="claude --dangerously-skip-permissions --resume"
TTS Volume
TTS notification volume is controlled independently of system volume via the VOLUME constant at line 25 of .claude/hooks/notification.py:
# Volume level: 0 (silent) to 1000 (max). Default 500 = 50%.
VOLUME = 400
Range: 0 (silent) to 1000 (max). Default 400 (~40%). Applies to both Windows MCI playback and ffplay/mpv fallbacks.
Codex CLI
Setup
Paste into Codex — or follow the steps manually:
Prerequisites: Node.js 18+, uv (python package manager), ffmpeg or mpv (audio for TTS).
1. git clone https://github.com/DefaultPerson/agent-setup.git && cd agent-setup
2. cp -r .codex/hooks ~/.codex/hooks
3. cp .codex/hooks.json ~/.codex/hooks.json
4. cp .codex/config.toml.sample ~/.codex/config.toml
5. cp AGENTS.md ~/.codex/AGENTS.md (optional — author's coding style and rules)
6. Edit ~/.codex/config.toml — set API keys, model preferences
7. Add MCP servers (see below)
8. Add shell aliases (see below)
9. Open `/hooks` in Codex and trust the PreToolUse + Stop hooks
10. Verify everything works
11. Delete agent-setup (repo no longer needed after setup)
Key differences from Claude Code:
- Config:
config.toml(TOML) instead ofsettings.json - Instructions:
AGENTS.mdinstead ofCLAUDE.md - Status line: built-in
/statuslinepicker, theme-aware colors viatui.status_line_use_colors, andtui.status_lineitems — no Claude-style custom script hook yet - Plugins: installed via interactive UI, not CLI command
- MCP: configured in
config.toml[mcp_servers]section or viacodex mcp add
Recommended Plugins
- agent-browser — browser automation for AI agents
- context7 — library documentation lookup
- frontend-design — production-grade frontend generation
Shell Aliases
Add to .bashrc / .zshrc:
alias cx="codex" cxr="codex resume" cxd="codex --yolo" cxdr="codex resume --yolo"
Tips
[!TIP] Disable desktop notifications — set
DESKTOP_NOTIFICATIONS=0in~/.claude/settings.json(envsection) or~/.codex/config.toml([shell_environment_policy].set). Audio TTS keeps working.
[!TIP] Terminal as Editor Tab (VS Code):
Cmd/Ctrl+Shift+P→ "Terminal: Create New Terminal in Editor Area" — opens terminal as a tab next to your code, not in the bottom panel.
[!TIP] If something doesn't work — just ask Claude Code/Codex to fix it. Describe the problem and it will diagnose and resolve it.
[!TIP] Create SKILLs for repetitive tasks. Instead of doing any task manually, create a SKILL for it. First version gives junior-mid level results. Then iterate until it matches your quality — 100-1000x time savings.
References
Inspiration for the rules and self-reflection patterns in CLAUDE.md / AGENTS.md:
- forrestchang/andrej-karpathy-skills — Behavioural principles for coding agents: Think Before Coding, Simplicity First, Surgical Changes, Goal-Driven Execution.
- DenisSergeevitch/chatgpt-custom-instructions — Rubric-driven self-evaluation methodology (model self-scores against a 5–7 category rubric and rewrites until threshold is met).
License
MIT