Work with the agent

The workspace screen is a tabbed panel: the primary AI tab is the agent’s terminal, the Git tab is a full changes/history/review view scoped to this worktree, and you can open any number of extra terminal tabs beside them. This page covers the day-to-day mechanics of steering an agent — answering its questions, feeding it files, and running your own commands next to it. For the bigger picture, see AI in Birch.

The tab strip

• AI tab — the agent’s terminal, always first. Its title updates to the agent’s own session title as work progresses.
• Git tab — Changes, History, Stash, and Review for this worktree, with a change-count badge showing files changed vs the base branch. See Review agent changes.
• Repo skills tab — the repository’s skills, when the Skills explorer is enabled.
• Extra terminal tabs — as many as you like, each closable with its ×.

Every terminal tab runs in the workspace’s worktree directory, not your main checkout. Background tabs stay fully alive when you switch away — the shell keeps running and the scrollback is intact when you return.

Answer questions and permission prompts

Agents pause to ask things — permission to run a command, a clarifying question. You answer in the terminal itself, exactly as you would in a standalone CLI session. What Birch adds is making the pause visible when you’re looking elsewhere: the workspace flips to the amber Waiting for your input status, which surfaces on the sidebar workspace row, the collapsed rail, the dashboard’s needs-attention list, an Inbox row (with an optional sound), and a desktop toast. Status detection is push-based and needs the birch-status plugin installed in your agent; notification behavior is configurable under Notifications. The full status vocabulary is decoded in Monitor status & usage.

Feed it files and links

• Drag & drop a file from your OS onto the terminal and Birch inserts its shell-quoted path at the cursor — the same behavior as a native terminal, and the fastest way to point an agent at a screenshot or log file. Dropping several files inserts all of their paths.
• Text and URL drags are pasted as text.
• http(s) URLs printed in the output are clickable and open in your system default browser — handy for dev-server addresses and PR links the agent prints.

Open extra tabs

The tab strip’s + button opens a menu:

• Terminal (⌘T / Ctrl+T) — a plain shell in the same worktree, for running tests, git commands, or a dev server beside the agent.
• Under AGENTS: Claude, Codex, OpenCode, and Custom command… — each entry shown only while enabled in Settings → AI → “New-tab harnesses” (the Custom entry also needs a non-empty Agent command). Agent tabs launch with your last-used model, effort, and permission mode.

There is no “Git” entry in this menu — the Git tab always exists and can’t be closed. Other shortcuts live in the keyboard shortcuts reference.

When the last terminal exits

When a workspace’s last terminal tab exits — you quit the agent, or close the final tab — the panel lands on the Git Review tab. That’s deliberate: the natural next step after an agent run is auditing what it did. See Review agent changes.

Open the worktree elsewhere

• The Path row in the workspace properties panel is clickable — it opens the worktree folder in Finder / Explorer — and has a copy button, so you can cd there from any terminal.
• For a terminal inside Birch, an extra Terminal tab already runs in the worktree — no navigation needed.
• A repository row’s context menu in the sidebar offers Show in Finder… / Show in Explorer…, Open in terminal, and Open in external editor for the repository itself (the editor is configured in Settings → General).