Skip to content
clandclandcland

Install

cland runs on macOS and Linux. Windows is not supported.

Install cland

Section titled “Install cland”

macOS: Homebrew

Section titled “macOS: Homebrew”
Terminal window
brew install cland

Linux: install script

Section titled “Linux: install script”
Terminal window
curl -fsSL https://cland.app/install.sh | sh

From source

Section titled “From source”

Requires Node 24.

Terminal window
git clone https://github.com/Kealman/cland-v0 cland
cd cland
npm ci
npm link

npm link puts the cland command on your PATH, pointing at the clone.

Check the install

Section titled “Check the install”
Terminal window
cland --version
cland doctor

cland doctor checks the harnesses, gh, Neovim, the hub and what cland init provisioned. It prints one line per check, ok, warn or fail, with the fix after — fix:. It exits 1 only when a check fails; a warning never fails it. Run it inside a project to include the project checks.

Requirements

Section titled “Requirements”
What Needed for
macOS or Linux Everything.
Git Every project. cland keeps the crew’s work in a local Git territory.
At least one harness Claude Code (claude), Codex (codex) or OpenCode (opencode). Each one a role uses must be installed and signed in.
gh, signed in Only when the project has a GitHub remote, and for importing GitHub issues.
A terminal The Neovim workbench and the harness logins. Any modern terminal works; Ghostty, kitty, WezTerm, iTerm2 and tmux are fine.
Node 24 Only for the source install.

You do not need Neovim installed or configured. cland downloads its own pinned Neovim release on first use, checks it against a pinned SHA-256 digest and starts it with a clean configuration, so your own Neovim setup is never loaded and never changed.

Harnesses and logins

Section titled “Harnesses and logins”

cland setup handles this for you. For each harness it:

  1. Looks for the binary, also in the official installers’ folders, and prints its version.

  2. If the harness is missing, shows its official user-level installer and runs it in your terminal when you say yes. A declined install is remembered and not offered again until you ask with cland setup --install <harness>.

  3. Runs the harness’s own login in your terminal, in the same environment the hub launches sessions with:

    Harness Login
    Claude Code claude auth login
    Codex codex login
    OpenCode opencode auth login --standalone (optional: the free model needs no login)

cland keeps OpenCode sessions in a database of its own, separate from OpenCode’s background service, so that service can never pick up a cland session. That is why the OpenCode login goes through setup: it signs in to cland’s database. cland doctor prints the exact login line for it.

Without a terminal, setup changes nothing: it prints each open step with its exact command and exits 1.

Terminal window
cland setup --json            # the steps as JSON
cland setup --install codex   # offer the Codex installer again
cland setup --no-accounts     # skip the accounts step
cland setup --no-aide         # do not launch the Aide

Harnesses update themselves. cland does not pin their versions; cland doctor prints each version for information.

Updating

Section titled “Updating”

Update cland the same way you installed it:

Terminal window
# Homebrew
brew upgrade cland


# Install script: run it again
curl -fsSL https://cland.app/install.sh | sh


# Source
cd cland && git pull && npm ci

After an update, restart each project’s hub so it runs the new version:

Terminal window
cland hub stop && cland hub start

cland doctor warns when a project’s hub runs another version than the CLI.

Updating a project’s skills and managed files

Section titled “Updating a project’s skills and managed files”

Each project carries its own copy of cland’s curated skill set in .agents/skills/, plus a few managed files (AGENTS.md’s cland block, docs/agents/, the Claude Code hooks, config.toml). Bring them to the installed version with:

Terminal window
cland update

cland update decides per skill:

  • A skill you never touched is refreshed.
  • A skill edited in the project is kept and reported as “edited in the project”. It belongs to the project now.
  • A skill new to the set is added, unless the project already has a folder of that name.
  • A skill dropped from the set is removed if untouched, and left to the project if edited.

It never rewrites roles.toml or crew.toml: those are yours. A file or hook without cland’s managed by cland mark is never written. Commit what cland update changed.

Uninstalling

Section titled “Uninstalling”

Remove cland itself:

Terminal window
# Homebrew
brew uninstall cland


# Source
npm unlink -g cland     # then delete the clone

For the install script, delete the cland command it installed (command -v cland shows where).

Remove cland from a project:

Terminal window
cland hub stop
cland init --undo      # restores the remotes: github becomes origin again

cland init --undo undoes only the territory step. It refuses when work would become unreachable and names the refs; keep that work under a surviving branch, or pass --force to discard it. Then delete .cland/, .agents/skills/, the .claude/skills link, the cland block in AGENTS.md and the cland hooks in .claude/settings.json if you no longer want them.

User-level state lives in:

Path What it holds
~/.config/cland/accounts.toml Your account names and paths. No secrets.
~/.local/state/cland/ The spend ledger of cloud verification runs.
~/.local/share/cland/opencode/ cland’s own OpenCode database.