Brainstorm → Build

Use the tool-free Brainstorm space to shape an idea into a spec, then hand the whole transcript to a coding chat with "Build this".

Brainstorm is a dedicated space for thinking before building. It’s a native Claude chat with no file tools — pure planning — and when you’re ready, “Build this” hands the entire conversation to a new coding-agent chat in a folder you choose.

💡 Brainstorm (no file changes) → 🔨 Build this → 🛠️ Coding chat in a folder → 🚀 Ship

The Brainstorm space

Brainstorm is a native Claude chat designed for one job: helping you think. Crucially, it has no file tools — it cannot read or write your code. That constraint is the point. There’s nothing to approve, nothing that touches your project, and no risk of the agent running off and editing files while you’re still figuring out what you actually want.

Use it to:

• Explore a vague idea and pressure-test it.
• Turn a one-line wish into a concrete spec.
• Compare approaches, list trade-offs, and decide on an architecture.
• Draft a step-by-step plan before any code exists.

Tip: Treat Brainstorm like a whiteboard session with a sharp collaborator. The messier and more open-ended your starting point, the more value you’ll get from talking it through here first.

How it differs from a coding chat

In short: Brainstorm plans, the coding chat builds.

”Build this” — the handoff

When your plan is solid, click “Build this”. Dices takes the whole Brainstorm transcript and hands it to a new coding-agent chat in a folder you choose. Your thinking carries straight over — the build agent starts with the full context of everything you discussed, instead of you re-explaining it from scratch.

1. Brainstorm until you have a plan you’re happy with.
2. Click Build this.
3. Choose the Folder (project) where the build should run.
4. A new coding chat opens, pre-loaded with the entire transcript, ready to implement.

Note: Because the build runs in a real Folder (usually a git worktree), this is where permissions, diffs, and commits come into play — exactly like any other coding chat. See The Cockpit and Quickstart.

History and expand

Two conveniences make Brainstorm comfortable for longer thinking sessions:

• History — a clock popover lets you get back to earlier brainstorm conversations, so a good thinking session isn’t lost.
• Expand to fullscreen — when you’re deep in planning, expand Brainstorm to fullscreen for a focused, distraction-free space.

Tip: Keep separate Brainstorm threads for separate ideas, and use the clock history to jump back to the one you want to turn into a build.

A worked example: idea → spec → build

Here’s the full loop in practice.

1. Idea (in Brainstorm)

You start vague:

I want a little web app where people can vote on lunch spots for the day.

2. Shape the spec (still in Brainstorm)

You and Claude talk it through — no files touched — and converge on something concrete:

Spec:
- Single page. List of lunch spots, each with an upvote button.
- Anyone can add a spot (name + optional link).
- Votes reset daily.
- Stack: a small static frontend + a simple key-value store.
- Show the current leader at the top.

3. Build this

You click Build this, pick your lunch-vote Folder, and a new coding chat opens with the entire spec conversation already loaded.

4. Build (in the coding chat)

The coding agent — running in the Folder’s git worktree — scaffolds the app, writes the code, and shows you a diff. You review it, ask for a tweak (“make the leader badge gold”), and the agent revises.

5. Ship

When the diff looks right, you commit:

Commit this with a clear message.

That’s the Brainstorm → Build arc end to end: think freely with no risk, then hand a complete plan to an agent that turns it into running, reviewable, committable code.

Next steps

• The Cockpit — where your build chats live and how sessions persist.
• Quickstart — the mechanics of reviewing diffs and committing.
• Introduction — the bigger picture and the full agent lineup.