Ultraplan runs in research preview on Claude Code v2.1.91 or later, and its behavior may shift as users weigh in. It breaks planning into three phases: targeted feedback, hands-off drafting, and flexible execution. You can flag specific sections for comments, keep your local terminal free while the plan’s drafted in the cloud, then either approve it to run on the web or bring it back to your CLI. From your local shell you kick off Ultraplan three ways: by running “/ultraplan” plus your prompt, by dropping the keyword ultraplan anywhere in a normal prompt, or by choosing “No, refine with Ultraplan on Claude Code” when a local plan finishes. In all cases, the /tasks command lists your active Ultraplan sessions with status symbols: ◇ ultraplan means Claude is drafting, ◇ ultraplan needs your input signals a question, and ◆ ultraplan ready tells you the plan awaits review. Stopping it archives the cloud session and clears the indicator without saving anything locally. Once the plan hits ◆ ultraplan ready, you open the session link in claude.ai to see the draft in a special review interface. You can highlight text and leave inline comments, tap emoji reactions, or jump between sections via the sidebar. When it’s polished, choose between two execution paths in your browser: “Approve Claude’s plan and start coding” to keep working in the cloud, or “Approve plan and teleport back to terminal” to fetch the plan into your local session. If you teleport back, your terminal shows an Ultraplan approved dialog with options to implement in place, start a fresh conversation, or cancel and save the plan to a file.