Launch CLED

CLED is a long-running daemon that must run in the user's GUI terminal session, so launch it in a dedicated iTerm2 window that outlives this turn. Run the preflight in order; stop with the specific fix if any check fails — don't launch a daemon that can't work.

Preflight

1. macOS — uname -s must print Darwin (agent detection uses iTerm2 + AppleScript). Else stop.
2. OpenRGB installed — osascript -e 'id of app "OpenRGB"' must succeed. Else offer brew install --cask openrgb and stop.
3. A controllable keyboard — if nc -z 127.0.0.1 6742 fails, start the server as CLED does: open -a OpenRGB --args --server --noautoconnect, then poll the port (up to ~30s; HID detection on macOS takes ~10s). Then /Applications/OpenRGB.app/Contents/MacOS/OpenRGB --client 127.0.0.1:6742 --list-devices must show a device with Type: Keyboard and Direct in its Modes:. Report the model; else stop — the keyboard isn't OpenRGB-controllable.
4. iTerm2 — osascript -e 'id of app "iTerm2"' must succeed. Else offer brew install --cask iterm2 and stop.

Launch

Open a dedicated iTerm2 window running the daemon (it persists in the GUI session, independent of this Claude session):

osascript <<'END'
tell application "iTerm2"
  set w to (create window with default profile)
  tell current session of w to write text "uvx cled"
end tell
END

uvx cled fetches CLED from PyPI and starts it. (Before CLED is published to PyPI, use uvx --from git+https://github.com/latent-spaces/cled cled instead.)

Confirm

Tell the user CLED is now running in a new iTerm2 window — number row = agent tabs (red busy / green idle / amber stale / blue other), F-keys = RAM, numpad = per-core CPU, Enter = heartbeat. To stop it: Ctrl-C in that window (OpenRGB keeps running).