Working on Parallel Tasks

A feature request lands. You want an agent on it now, but you’re in the middle of something and not about to drop it. Here’s how that plays out in Loxel.

Create a worktree for the task

Press Control+Option+NN to open the New worktree dialog. Type a name — something like user-invites — and press ReturnEnter. Loxel creates the worktree and its branch with wt, and it appears in the sidebar. Click it to switch to the new task.

The task you left is saved as it was: same files open, same terminals, same layout.

Hand off the task

With the new task active, press Command+TT to open a terminal. It starts in the task’s worktree, so start Claude Code, Codex, or whatever agent you use, and tell it what to build. See Running Coding Agents for notifications and the loxel skill.

Once the agent is running you don’t need to stay here.

Go back to what you were doing

Press Control+Option+[[ to go back to the task you came from. The layout snaps back — same open files, same panel arrangement. The agent keeps working in the background.

You’re unblocked. Get on with your day.

Come back when the agent is done

A dot on a worktree in the sidebar means one of its terminals raised a notification, and the commit graph shows an uncommitted-changes row for each worktree with pending changes. When you see work piling up in the task, or just want to check in, press Control+Option+]] to switch back.

Switch to the agent’s terminal tab to read what it said.

Review the diff

Open the Changes panel (Command+Shift+CC) to see what the agent produced: every changed file, with added and removed line counts. Double-click a file to open it in the diff viewer — side-by-side by default, with synchronized scrolling, gutter connectors between the two sides, and the changed characters highlighted within each line. See Diff Viewer for details on modes.

Leave comments

Open the Comments panel, create a review session (click Reviews in the header → new review icon → name → ReturnEnter), and go through the diff file by file. Select any span of code that raises a question and click Add comment in the gutter. Comments are anchored to content, not line numbers — they follow the code if the agent rewrites it. Leave intent-based questions (“is this the right abstraction boundary?”) rather than line-level corrections. See Code Review for the full anchor system reference.

Iterate if needed

Send a follow-up message to the agent in its terminal. Comments track through the rewrite. When the agent finishes, outdated comments keep the original lines under “Original code at time of comment” in the Comments panel — compare them with the new code to verify whether your concern was addressed.

Commit, done

Once you are satisfied, commit from a terminal in the task (or ask the agent to). Your other tasks never noticed.