Felipe Frizzo's dotfiles for OSx

This repository include all of my custom dotfiles.

This include the following step.

• install xcode-cli
• install homebrew
• install all applications from homebrew
• install atom plugins
• install external softwares
• run set OsX system defaults
• run set user defaults
• copy bash configurations to home user folder
• install latest updates from app store
• configure Dock with the applications
• link AI agent instructions (Claude Code / Codex / Copilot) to a single source
• install third-party agent skills
• install the caveman skill for every AI agent found on the machine
• install graphify (uv tool + skill)
• install the iTerm2 dynamic profile

Installation

Download

mkdir dotfiles
cd dotfiles
curl -#L http://github.com/felipefrizzo/dotfiles/tarball/master | tar -xvz --strip-components 1
chmod +x *.sh
sh setup.sh
cd ..
rm -rf dotfiles

Also run dotfiles-confidential's setup.sh for personal casks/formulas and .ssh/config that don't belong in a public repo.

AI agent config (Claude Code / Codex / Copilot)

setup.sh tracks and restores:

• the hand-authored parts of ~/.claude/settings.json and ~/.copilot/settings.json
• a minimal ~/.codex/config.toml fragment (model, model_reasoning_effort, shell_environment_policy.inherit) -- the rest of that file is written by the ChatGPT desktop app itself and regenerates from normal use, so it's intentionally not tracked
• ~/.agent-instructions/AGENTS.md (tool-agnostic rules), symlinked into Codex's AGENTS.md, Copilot's copilot-instructions.md, and ~/.claude/AGENTS.md directly
• ~/.agent-instructions/CLAUDE.md (Claude-Code-specific model/subagent routing), symlinked into ~/.claude/CLAUDE.md. It imports @AGENTS.md, which needs ~/.claude/AGENTS.md to exist as a sibling file since @file import resolution isn't guaranteed to follow the symlink to its real path -- hence the separate ~/.claude/AGENTS.md symlink above
• ~/.agent-instructions/RTK.md, symlinked into both tools' RTK.md -- edit once, every tool sees it; the hook that auto-rewrites shell commands through rtk is Claude-Code-only, other tools invoke rtk directly
• third-party agent skills, via home/skills-manifest.txt (one npx skills add <source> per line, regenerate from ~/.agents/.skill-lock.json)
• the caveman skill, via its own installer
• graphify, via uv tool install graphifyy then graphify install --platform claude|codex -- the package ships its own skill file and installs it itself, so it's intentionally not hand-copied into this repo (would drift from whatever graphifyy ships on the next version bump)

Not tracked, ever: API tokens/credentials of any kind. GITHUB_TOKEN and JIRA_TOKEN (used by the rtk hook and Codex's shell environment policy) must be set up manually on each machine -- they are intentionally absent from every file this repo manages, public or confidential.