dev-stack

A unified development environment: a provisioning script plus the actual dotfiles it manages, for macOS, Red Hat, and Ubuntu. Clone this repo anywhere and run setup-env.sh to get the same nvim config, tmux setup, and project launcher on a new machine.

MIT licensed, see LICENSE.

• macOS: Homebrew must already be installed. If it isn't, setup-env.sh just warns and skips package installation entirely, nothing else in the script depends on Homebrew, so dotfile linking and dev-stack still work.
• RHEL / Ubuntu: dnf/apt-get, plus sudo access to install system packages. Without sudo, see No sudo access.
• git and bash on all three.

setup-env.sh    Provisioning script - installs packages, links dotfiles below
nvim/           LazyVim config (dotfile source of truth, symlinked to ~/.config/nvim)
tmux/tmux.conf  tmux config (dotfile source of truth, symlinked to ~/.tmux.conf)

nvim/ and tmux/tmux.conf are the real, live config files, not templates. Edit them here, commit, and re-run setup-env.sh on any machine to pick up the change (or just reload tmux/nvim directly, since the symlink means the edit already took effect).

setup-env.sh is idempotent, safe to re-run any time:

• Detects your OS (macOS / RHEL / Ubuntu) and installs the right packages with the right package manager (brew, dnf, apt-get).
• Follows the Perl Principle, applied equally to Python: never forces a system-wide Perl or Python reinstall if a working binary is already on PATH (important if you manage either with something like perlbrew, pyenv, or conda, since the OS package manager installing "on top" can silently change what perl/python3 resolve to, or even overwrite existing symlinks via brew link). Only installs Perl::Tidy and Perl::Critic via cpanm if they're missing, since nvim/lua/plugins/perl.lua already wires up perlnavigator, perltidy, and perlcritic.
• Installs global web tooling (prettier, live-server) to a user-owned npm prefix, not root.
• Symlinks ~/.config/nvim to this repo's nvim/, and ~/.tmux.conf to tmux/tmux.conf. If a real (non-symlink) config already exists at either location, it's renamed to a timestamped .bak next to it first - never deleted, so a first run on a machine with its own hand-built config is safe; nothing is lost, you just end up with a .bak directory to compare against and discard once you're happy.
• Keeps tmux on the factory-default Ctrl+b prefix intentionally, so it never collides with LazyVim's Ctrl+h/j/k/l pane navigation from vim-tmux-navigator, plus installs tmux-persist, tmux-yank, and extrakto.
• Installs ~/.local/bin/dev-stack, a project launcher (see below), with dev-start kept as a symlink to it.
• Adds shell completion for dev-stack to both ~/.bashrc and ~/.zshrc (dev-start is kept for compatibility only, without completion).

Windows is not yet supported. See the stub at the bottom of the script.

One-liner, no manual clone needed:

curl -fsSL https://raw.githubusercontent.com/rshingleton/dev-stack/main/install.sh | bash

install.sh is the only thing safe to pipe into bash sight-unseen (it does nothing but clone/update the repo to ~/dev-stack and hand off); everything else - packages, symlinks - lives in setup-env.sh, which install.sh runs for you. Re-running the one-liner later just pulls the latest and re-runs it, same idempotent behavior as running it locally.

Override where it's cloned or which repo it pulls from with env vars:

DEV_STACK_DIR=~/code/dev-stack DEV_STACK_REPO_URL=https://github.com/rshingleton/dev-stack.git \
  curl -fsSL https://raw.githubusercontent.com/rshingleton/dev-stack/main/install.sh | bash

Or clone it yourself and skip the bootstrap entirely:

git clone https://github.com/rshingleton/dev-stack.git ~/dev-stack
cd ~/dev-stack
chmod +x setup-env.sh
./setup-env.sh

setup-env.sh finds its own location (SCRIPT_DIR), so both paths behave identically, there are no hardcoded machine-specific paths. If you're setting this up for yourself rather than forking it, you'll likely want to replace nvim/ and tmux/tmux.conf with your own config before running it, the LazyVim setup here (Perl LSP, debugger, test runner) is tailored to a Perl-centric workflow.

dev-stack launches (or reattaches to) a tmux session for a project, with nvim, Claude Code, and a small shell all laid out for you. No project list is hardcoded in the script; aliases are registered per machine. dev-start is kept as an alias, so every command below works under either name.

dev-stack init <name> <path>   Register an alias for a project path
dev-stack list                 Show registered aliases
dev-stack status               Show which aliases are running, restorable, or neither
dev-stack pick                 Fuzzy-pick a registered alias and launch it (uses fzf if installed)
dev-stack remove <name>        Remove an alias
dev-stack export [file]        Print (or save) aliases, to move to another machine
dev-stack import <file>        Merge aliases from a file (fix paths afterward)
dev-stack [name-or-path] [session-name]

Aliases live in ~/.config/dev-start/projects, a machine-local file (project paths differ from machine to machine, so this is deliberately not synced or committed). Use export/import to move the list between machines and fix the paths by hand.

Each session gets one window with nvim, Claude Code, and an idle shell for git/cmd-line ops:

• Vertical (default): nvim on top, Claude Code in the middle, shell sliver at the bottom.
• Horizontal: nvim on the left (full height), Claude Code top-right, shell sliver bottom-right.

Set DEV_START_LAYOUT=horizontal to switch, e.g. DEV_START_LAYOUT=horizontal dev-stack myproject.

dev-stack also knows about tmux-persist: since persist auto-restores a session's saved layout the instant a session with a matching name is created, dev-stack checks for an existing snapshot first and steps out of the way (just creating a bare session) rather than racing persist's restore with its own layout-building commands.

chmod +x setup-env.sh
./setup-env.sh

Re-run any time; every step checks before it acts.

On machines where you can't install system packages yourself, setup-env.sh still does everything that doesn't need root (dotfile links, dev-stack, completion, npm tools, tmux plugins) and hands the rest off instead of failing:

• It checks once up front: root, passwordless sudo, or a sudo password prompt if there's a terminal. With no way to get sudo (or when Homebrew on macOS is owned by another user), missing packages are collected rather than installed.
• At the end it prints the exact install command(s) and writes them to ~/dev-stack-prereqs.sh. Hand that to an admin to run as root (sudo bash ~/dev-stack-prereqs.sh), or on macOS to the user that owns Homebrew.
• Re-run ./setup-env.sh once they're installed to finish the setup. The hand-off file is removed once nothing is missing.
• If the system Perl's library isn't writable, Perl::Tidy and Perl::Critic are installed into ~/perl5, with PATH/PERL5LIB added to ~/.profile.

The hand-off installs one package at a time, so one the repos don't carry doesn't block the rest; it reports any that failed. On RHEL, neovim, fd-find and ripgrep come from outside the base repos, and the hand-off notes that EPEL needs enabling first.

lazygit never needs root on RHEL/Ubuntu (it isn't in EPEL or every Ubuntu release): it's built with go install into ~/.local/bin when Go is on PATH, and otherwise (or if that fails) the checksum-verified release binary is installed there instead.

Options:

./setup-env.sh --no-sudo          Don't attempt installs; always produce the hand-off (or DEV_STACK_NO_SUDO=1)
./setup-env.sh --print-prereqs    Only print what's missing, change nothing

Both pass through the one-liner too: curl -fsSL .../install.sh | bash -s -- --no-sudo.