If you're working inside an Orca editor worktree, this is how you talk to Orca's control plane instead of running raw git worktree commands or spawning ad hoc shells. You get JSON commands to create worktrees per issue, update the worktree comment at meaningful checkpoints so the UI shows your progress, and read or write to Orca-managed terminals. The boundary that matters: use this for non-agent terminals like build scripts or test runners, but switch to the orchestration skill when the recipient is another AI agent. It keeps worktree state consistent and gives you a single handle for terminal lifecycle, which beats managing PTYs yourself.
npx skills add https://github.com/stablyai/orca --skill orca-cli