How to use the brainstorming agent skill: install and trigger it

What the brainstorming skill makes the agent do

An agent skill is a folder of instructions that an AI agent loads when a task matches it. The brainstorming skill applies to creative work: building a feature, adding a component, changing behaviour, or planning something new, whether that is software or something else such as an event, a shop or a renovation.

Instead of building straight away, the agent first works out what you want and why. It asks one open question at a time, offers sensible defaults that it states plainly, and can suggest a quick throwaway prototype when that is the cheapest way to learn. For larger projects it writes a short design document and waits for your approval before any building starts.

• Asks open questions about the purpose, the success criteria and what you have already tried
• Restates its understanding in short summaries and asks what is wrong or missing
• Sizes the work: quick and clear tasks are done directly, larger ones get a written design
• Writes a design document for projects, then has a fresh sub-agent read it for unanswered questions
• Does not build anything until you approve the design (a hard gate in the skill)

Install the brainstorming skill

Download the ZIP of the brainstorming skill from this page. Then add it to the agent in one of two ways, depending on where you use it.

For Claude Code, unzip the folder into ~/.claude/skills/ to make it available in all your projects, or into a project's .claude/skills/ folder to limit it to that project. In the Claude apps, upload the skill under Settings → Capabilities → Skills.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into the project's .claude/skills/ folder
• Claude apps: Settings → Capabilities → Skills, then upload the skill

Trigger the skill with natural requests

You do not need a special command. The skill is meant to apply when you ask for something creative or non-trivial. Describe what you want in ordinary words, and the agent decides whether to brainstorm first.

If you want to skip the questions, say so in your first reply. The skill then ends and the agent carries out your request in its normal workflow.

• "I want to add a way for users to export their data. Help me think it through before you code."
• "I'm planning a small neighbourhood sharing evening. Ask me what matters before suggesting anything."
• "We're renovating the kitchen. Let's plan it before buying anything."
• "Just go ahead and build it, I don't need questions."

Example: a planning session from start to finish

Suppose you ask for a booking page for a bakery. The agent first asks when you would have wanted a tool like this, and who uses it. As you answer, it summarises the parts it understands and asks what it has missed. Where a choice is yours to make, such as the platform or language, it proposes a default and says so.

Once the picture is clear, the agent writes a design covering intent, goals, non-goals, constraints and what is left for the builder to decide. You read it, correct it, and approve it. Only then does the agent move on to planning and building.

What the skill needs and the safety notes

The skill includes scripts. Its browser helper, helper.js, is written to run in browsers and in Node. The start and stop scripts for the optional visual companion are Bash scripts, so they need a Bash shell. Node.js or a compatible runtime is needed for the helper and tests, and a browser is needed to view the visual companion.

The skill does not use the network, according to its listing. The visual companion runs a local server that watches a folder of HTML files and shows the latest one in your browser. Read the scripts before you run them, and start the companion only after you agree to use it.

• Runs scripts: yes (start-server.sh, stop-server.sh, helper.js)
• Uses network: no, according to the listing
• Requires Node.js or a compatible runtime, a Bash shell and a browser for the optional visual companion

Tips for getting the most from it

Give the agent a real moment or example rather than a general wish. Answers about specific events tend to produce better designs than abstract goals.

Use the visual companion only when seeing something helps, such as a layout, a flow or side-by-side options. Keep requirements and trade-offs in text.

• Answer in your own words; the agent will offer options only when you get stuck
• Correct its guesses. Guesses it labels as guesses can be fixed quickly
• Read the design document before approving it; approval covers only what you were shown
• If you start a prototype, treat it as throwaway and expect the agent to brainstorm again before keeping it