The problem with planning in the CLI is that you are stuck in a linear, terminal-bounded loop.
You run/plan, read a wall of text, approve or reject the whole thing, and move on.
There is no way to comment on a specific section, comment on something you agree with, or flag something you want changed without rewriting the entire prompt.
/Ultraplan fixes this.
You run commands from your local CLI, and the cloud generates the implementation plan remotely on Anthropic’s cloud infrastructure.
When this happens your terminal remains free.
When the plan is ready, you open it in a browser on claude.ai and interact with it like a document: inline comments on specific passages, emoji reactions per section, and an outline sidebar to move around.
You can ask the cloud to modify, iterate, and redraft until the plan feels right.
Once approved, you choose: execute the plan in the same cloud session and receive a PR, or teleport it back to your terminal with full local environment access.
What makes it interesting is the divide between where the planning happens and where the execution happens.
Planning is mostly reading and reasoning, so the cloud is fine.
Execution often requires your local environment. /Ultraplan respects that distinction.
what’s that for: Engineers and teams already using cloud code require more control at the planning stage, before the code is written.
Especially useful for large refactors, migrations, or anything where blindly approving a plan feels risky.
thing worth noting: Requires Cloud Code v2.1.91 or later, Cloud Code enabled on the web, and a GitHub repo. Not available on Bedrock, Vertex or Foundry.
<a href