Terminal

Coldtea's terminal is a real terminal pane for local CLI agents and shell commands.

That is the main product boundary: Coldtea hosts the agent instead of replacing the agent with a custom chat transcript.

What real terminal means

The agent runs as a normal process in a PTY. Its output, prompts, permission requests, colors, and interactive behavior belong to the CLI you launched.

If the agent asks you to approve a command, you approve it in the terminal. If it needs login, complete the CLI's login flow. If a command is dangerous, stop it the same way you would outside Coldtea.

Coldtea adds the surrounding workspace: panes, tabs, brews, tasks, editor context, worktrees, search, and annotations. It does not make terminal commands safe by being nearby.

Starting from an empty pane

An empty terminal pane shows installed agents and quick-launch shortcuts. Use Ctrl+1 through Ctrl+9 to start the first installed agents from the workspace home.

You can also open the composer with Cmd+N or Ctrl+N, choose an agent, and send the first prompt.

Panes, tabs, and splits

Use tabs for separate lanes of work. Use splits when you need two panes visible at the same time, such as an agent beside a dev server or a test command.

Common shortcuts:

  • Cmd+T / Ctrl+T creates a new terminal tab.
  • Cmd+W / Ctrl+W closes the active pane or tab.
  • Cmd+D splits right on macOS.
  • Cmd+Shift+D splits down on macOS.

Check the active path and branch after switching panes. A focused terminal is the one that gets your keystrokes.

Search, clear, copy, paste

Use Cmd+F or Ctrl+F to search the active terminal. Use Cmd+K on macOS to clear the visible pane.

Coldtea forwards common terminal editing shortcuts into the PTY. The active shell or agent can still capture some keys first, so use the visible terminal behavior as the source of truth.

Selecting text lets you copy it or turn it into review context where annotation actions are available.

Comments and annotations

Terminal annotations are useful when the important context is output, not code. Select or click the relevant terminal line, add the comment, then send the comment back to the agent when you want it addressed.

Do not assume the agent automatically saw your comment just because it exists in the UI. Send or attach it through the available action, or quote the relevant terminal output in the prompt.

A safe terminal habit

Before running or approving a command, check three things:

  1. Which project or worktree is this pane in?
  2. Which agent or shell process is active?
  3. What would this command touch if it is wrong?

That habit is slower than pressing Enter. It is also cheaper than cleaning up the wrong checkout.

Next: start or continue agent work or learn how agents run in Coldtea.

On this page