Skip to content

Guide · Agent skill

How to use the Brainstorming Before Building agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

The brainstorming skill from obra/superpowers changes how an AI agent handles new work. Instead of starting to build as soon as you ask for something, the agent works out what you want and why, asks focused questions, and waits for your approval of a short design first. This guide explains what the skill does, how to install it, how to trigger it, and where it is a poor fit, so you can decide whether to use it.

Brainstorming Before Buildingobra/superpowers · MIT

Quick start

  1. 1Download the ZIP. Download the brainstorming skill ZIP from this page.
  2. 2Unzip the skill. Unzip the folder into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder for one project.
  3. 3Or upload it in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the skill.
  4. 4Make a non-trivial request. Describe what you want to create or change, in plain words, including anything you are unsure about.
  5. 5Answer the focused questions. Reply to the agent's questions. It will reflect back what it understood and propose approaches.
  6. 6Approve the design. Choose an approach and approve the short design before the agent starts writing code.

What the brainstorming skill makes the agent do

When you ask the agent to create or change something, the skill tells it to check its own assumptions first. If your request already contains what matters, the agent does the task and says what it did. If something important is missing, it starts a conversation: it asks one focused question at a time, reads your existing project if there is one, and proposes a few ways to build the thing with trade-offs.

The skill requires that the agent lays out the approaches, you pick one, and you approve a written design before any implementation begins. Simple, clearly specified tasks skip this process.

  • Asks questions about what you want and why, instead of guessing
  • Reads the existing project or the material you point to before asking
  • Presents two or three approaches with trade-offs and a recommendation
  • Waits for your approval of a design before writing code
  • Can offer an optional browser view for mockups or visual options

Install the brainstorming skill from the ZIP

Download the ZIP for the brainstorming skill from this page and unzip it. The skill is a folder that contains a SKILL.md file. Put that folder in one of these locations, depending on how you want to use it:

  • Personal use across projects: unzip into ~/.claude/skills/
  • One project only: unzip into the project's .claude/skills/ folder
  • In the Claude apps: upload the skill under Settings → Capabilities → Skills

Trigger the skill with natural requests

You do not need a special command. Describe the thing you want to create or change, in plain words, and say what you are unsure about. The skill is written to run whenever the agent is asked to create or change something non-trivial. You can also ask for brainstorming by name, which tells the agent you want a design conversation rather than code.

If you want the agent to skip the questions, say so. The skill allows you to opt out and have the agent proceed with what you asked.

  • "Let's brainstorm a booking feature for my salon app before we build anything."
  • "I want to plan a renovation of my kitchen. Ask me what I need before suggesting anything."
  • "Help me think through a website for my bakery."
  • "Make the icon cornflower blue." (a quick task, so the agent will just do it)

Examples of a good session

For a bakery website, the agent does not ask for a list of pages. It asks what someone should do when they land on the site, because that answer decides whether the site shows hours and the day's products or takes orders. It then proposes a direction and waits for your correction.

In an existing project, the agent reads the relevant code first and reports how it works today and what your change would touch, so your questions focus on decisions the code cannot answer.

What you need before you start

The skill itself is a set of instructions for the agent. The optional visual companion is a set of scripts, and it has its own requirements. Read them before you turn it on.

  • Bash shell and Node.js, to run the optional visual companion scripts
  • A browser, to view the visual companion pages
  • The skill runs scripts, but it does not use the network

Safety notes for the visual companion

The optional visual companion includes scripts (scripts/start-server.sh and scripts/stop-server.sh) that serve a local web page showing mockups or options. The companion can bind to a network address, so do not start it on a shared or remote machine. Stop it with scripts/stop-server.sh when you finish.

You can skip the visual companion entirely. The rest of the skill works through conversation alone.

  • Do not run the companion on a shared or remote machine
  • Stop the companion when you are done
  • Review the scripts in the folder before running them

Tips for getting the most out of it

Answer the agent's questions in your own words. The skill is designed to learn from your descriptions, and your wording often carries intent that a menu of options would miss.

If the agent proposes a sensible default, you can accept it or correct it. When you say you want to just guess, the agent proposes something concrete instead of asking again.

  • Give context early: a spec, a plan or files you have already written
  • Say why you want the thing, not just what it is
  • Use the visual companion for layouts, flows and side-by-side options
  • Ask for a quick spike when you want to feel out an idea before committing to a design

What people use it for

  • Planning a new feature in an existing software project, with the agent reading the code first
  • A non-programmer thinking through a website or app before hiring anyone or writing a brief
  • Working out a business plan or a talk's structure through guided questions
  • Planning a home renovation by listing needs, constraints and trade-offs before buying materials
  • Reviewing a vague spec with the agent to find the open questions before building

Frequently asked questions

How do I install the brainstorming agent skill?

Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder for one project. In the Claude apps, you can upload it under Settings → Capabilities → Skills instead.

Do I need to know how to code to use it?

No. The skill is written for planning anything new, including websites, businesses, talks and renovations. The optional visual companion needs a Bash shell, Node.js and a browser, but the rest of the skill works through conversation.

Is the brainstorming skill free?

The source is licensed under MIT. Check the licence file in the repository for the full terms. This guide does not state any price.

Does the skill need an internet connection?

The skill's assessment says it does not use the network. The optional visual companion serves a local web page on your computer, and it can bind to a network address, so do not run it on a shared or remote machine.

Why does the agent keep asking questions instead of building?

The skill is designed to pause for conversation before building anything non-trivial. If you want code immediately, say so and the agent can proceed with what you asked. Simple, fully specified tasks like changing one colour are just done.

What are the alternatives if this skill is not right for me?

If you want code immediately and find questions slow, skip this skill and give the agent a fully specified request. For visual planning without the browser companion, you can ask the agent for a text-based comparison instead.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you are starting a new feature or planning something non-technical and want the agent to check assumptions before building the wrong thing. Pick something else if you mostly give quick, fully specified tasks or do not want to answer questions, and avoid running the visual companion on a shared or remote machine.

Download the skill