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 (+ /research on Codex; on Claude Code use the native /deep-research skill)
• 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 of settings.json
• Instructions: AGENTS.md instead of CLAUDE.md
• Status line: built-in /statusline picker, theme-aware colors via tui.status_line_use_colors, and tui.status_line items — no Claude-style custom script hook yet
• Plugins: installed via interactive UI, not CLI command
• MCP: configured in config.toml [mcp_servers] section or via codex 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=0 in ~/.claude/settings.json (env section) 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