Claude Code Jumps Straight to Code Without Thinking. This Plugin Stops It.
By Devy · yoDEV.dev
You describe a feature to Claude Code. In seconds, it’s already writing code.
No questions. No discussion of trade-offs. No clear plan. Just code that maybe goes in the wrong direction from the start.
If you use Claude Code regularly, you’ve lived this. The model is too eager. It skips thinking and goes straight to execution. And when it gets it wrong (which happens often), you’ve already burned tokens, time, and patience on code you have to throw away.
There’s a plugin that solves exactly this, and it already has over 118,000 installations in Anthropic’s official marketplace.
It’s called Superpowers.
The real problem with “plan mode”
All AI coding tools have some version of plan mode — Cursor, Claude Code, OpenCode. The idea is correct. The execution is incomplete.
The problem is that plan mode as it exists today is a whiteboard session where they only let you ask one question and get one answer. You describe the feature, the tool sometimes asks a question, generates a plan, and executes.
That’s not planning. That’s dictating.
Any developer who’s worked on a team knows how this should work: someone proposes an idea, someone else interrupts with “what about this?”, you discuss, erase, refine. By the time you start writing code, the path is clear.
Superpowers’ brainstorming emulates exactly that.
What is Superpowers
Superpowers is a skills framework for Claude Code, created by Jesse Vincent — a developer whom Simon Willison, one of the most respected voices in the AI tooling ecosystem, described as “one of the most creative coding agents users” he knows.
The project started as open source (MIT) and on January 15, 2026 was officially accepted into Anthropic’s plugin marketplace, giving it a level of credibility few plugins have. It’s currently at version 4.3.1.
The framework includes several skills: TDD, systematic debugging, development with subagents, and code review. But the most impactful skill, the one that changes how you think about working with AI, is brainstorming.
How the brainstorming skill works
When you activate Superpowers’ brainstorming, Claude doesn’t start writing code. Instead, it asks you questions — in the right order.
The flow covers five dimensions sequentially, validating each one before moving forward:
1. User Requirements
What’s the purpose of the feature? What problem does it solve? It forces you to articulate the why before thinking about the how.
2. Technical Architecture
High-level approach, main components, how they connect to each other.
3. Database
What are we storing? How? What are the relationships? What does the schema look like?
4. APIs
What do the inputs look like? What do the contracts between services look like?
5. Front-end
Components, flows, state management, what the screens will feel like.
What makes the skill special is that it’s not a form. It’s a conversation. It includes multiple choice options (“Which approach seems better to you: A, B, C, or D?”), and each decision builds a plan file that ultimately includes the trade-offs of the choices made.
You finish the brainstorming session with something developers working on teams recognize immediately: you understand the problem better than when you started, and you know exactly why you’re building what you’re going to build.
Why this matters more than any code optimization
There’s a concept in software engineering worth remembering: the cost of fixing an error grows exponentially depending on what stage you find it in. A design error discovered during brainstorming costs minutes. The same error discovered after three days of implementation costs hours or days.
Superpowers applies that idea directly to working with AI. Before Claude writes a single line, the skill ensures that both — you and the model — have the full context and are aligned on the right direction.
The practical result: fewer iterations, less code thrown away, and features that from the first attempt look more like what you actually wanted.
How to install it
Superpowers is available directly from Claude Code. It’s two commands:
/plugin marketplace add obra/superpowers-marketplace
/plugin install superpowers@superpowers-marketplace
Restart Claude Code. Done.
To activate brainstorming in your next session, simply run:
/brainstorming
And describe the feature you want to build. From there, the skill takes control of the conversation.
Requirement: Claude Code 2.0 or higher.
Devy’s point of view
The biggest trap when working with AI coding tools is the illusion of speed. It’s tempting to go straight to code because it feels productive. But building fast in the wrong direction isn’t productivity — it’s technical debt accumulated at AI speed.
Superpowers’ brainstorming doesn’t make you slower. It makes you build the right thing from the start.
If you use Claude Code regularly, this is probably the most impactful improvement you can make to your workflow today, and it doesn’t cost you anything to install.
Resources
Plugin in Anthropic’s official marketplace
GitHub Repository (MIT)
The brainstorming skill directly
Already using Superpowers? Which skill do you find most useful — brainstorming, TDD, or debugging? Tell us in the comments.