/ultraplan - Plan Complex Tasks in a Cloud Session
Use /ultraplan to plan in a cloud session while your terminal stays free. Review, comment, and iterate on the plan in your browser.
What is it
When you ask Claude Code to plan something big (a new feature, a refactor across twenty files, an architecture decision), it normally does that thinking right in your terminal. That works fine for quick plans, but for anything that needs real thought, your terminal is locked up while Claude researches your codebase and drafts the plan.
/ultraplan moves that planning work to a cloud session. Claude spins up a separate Opus 4.6 session with up to 30 minutes of compute time, does all the research and planning there, and you get a rich browser-based review when it's done. Your terminal stays free the whole time.
How /ultraplan compares to /plan and ultrathink
These three sound similar but do completely different things. Here's the quick breakdown:
/plan plans locally in your terminal. Claude reads your codebase, drafts a plan, and shows it to you right there. Simple and fast, but your terminal is busy the whole time.
/ultraplan plans in the cloud. Claude does the same research and drafting, but in a separate cloud session. You review the plan in your browser with inline comments and reactions. Your terminal stays free.
ultrathink is not a planning command at all. It's a keyword you drop into any prompt to make Claude think harder on that one response. It bumps the reasoning effort to "high" for a single turn, then goes back to your default. Works on Opus 4.6 and Sonnet 4.6 only.
| What it does | Where it runs | Your terminal | Context window impact | |
|---|---|---|---|---|
/plan | Drafts a plan | Your machine | Blocked | Eats your local tokens |
/ultraplan | Drafts a plan | Cloud session | Free | Separate budget, yours stays clean |
ultrathink | Thinks deeper | Your machine | Blocked | Uses more thinking tokens locally |
Why the token difference matters
This is the part most people miss. When you run /plan, every file Claude reads, every search it runs, every piece of your codebase it explores... all of that fills up your local context window. By the time the plan is done and you say "go build it," you've already burned through a chunk of your available tokens. Claude has less room to work with during the actual coding.
With /ultraplan, all that research happens in a separate cloud session with its own token budget (up to 30 minutes of compute). Your local context stays completely untouched. When the plan is done and you teleport it back to your terminal, you start the implementation phase with a full, clean context window. More room for Claude to read files, write code, and run tests without hitting the limit.
Think of it this way: /plan is like drafting blueprints and building the house on the same desk. /ultraplan drafts the blueprints at a separate office and brings them to your clean workspace when they're ready.
ultrathink keyword has nothing to do with planning. Use it when you need Claude to reason harder about a tricky bug or a complex decision. Add the word "ultrathink" anywhere in your prompt. It temporarily sets effort to high for that one response.What problems it solves
You want to add a new feature to your app, say a dashboard page with charts, filters, and user settings. There are existing components to reuse, a data layer to connect to, and routing to set up. You need Claude to look at your whole project, figure out what's already there, and come up with a plan that fits. That's 10-15 minutes of thinking, and your terminal is just sitting there the whole time.
Or you're starting a brand new app from scratch. You tell Claude "I want to build a habit tracker with streaks, reminders, and a calendar view" and the plan it comes back with is massive. You want to say "skip the reminders for now" or "use a simpler layout for v1." In the terminal, that's a clunky back-and-forth. In the browser, you highlight the section you disagree with and type your feedback right there.
How to use it
Launch it with the command:
/ultraplan add a dashboard page with analytics charts and date filtersYou can also just drop "ultraplan" into a normal prompt and Claude picks it up:
I want to build a recipe sharing app with search and favorites, ultraplan this for meA confirmation dialog pops up. Once you approve, your terminal shows a status indicator while the cloud session works:
◇ ultraplan Claude is researching and drafting
◇ ultraplan needs your input Claude has a question, open the link
◆ ultraplan ready Plan is ready to review in browserRun /tasks to see the ultraplan entry and grab the session link.
Reviewing in the browser
When the plan is ready, you open it in your browser and get:
- Highlight any section and leave inline comments
- Drop a thumbs-up or flag on sections you want changed
- Jump between sections using the outline sidebar
- Ask Claude to address your comments, then review the next revision
Executing the plan
Once you're happy with it, you have two choices:
Run it in the cloud. Click "Approve Claude's plan and start coding" in the browser. Claude implements the plan right there. Create a PR from the web interface when it's done.
Send it back to your terminal. Click "Approve plan and teleport back to terminal." Your CLI shows the plan with three options: implement it in your current session, start a fresh session with only the plan as context, or save it to a file.
Pro tips
Start with /ultraplan for anything that needs more than 5 minutes of planning. If Claude can plan it in under a minute, local planning in the terminal is faster. The cloud session has overhead: spinning up, switching to the browser, reviewing there. It only pays off when the planning itself is substantial.
The "teleport back to terminal" option is my preferred flow. I like the browser review for feedback, but I want execution happening locally where I can watch the diffs and run tests immediately. Best of both worlds.