How to use the New App Builder agent skill (new-app)

What is an agent skill and what does new-app make the agent do?

An agent skill is a folder containing a SKILL.md file with instructions that an AI coding agent loads when a request matches the skill's description. The new-app skill applies to requests to create a new application, project, website, game, mobile app, CLI tool or library from scratch.

Once triggered, the skill makes the agent follow a fixed sequence rather than starting to write code immediately. It first asks clarifying questions, then proposes a plan and waits for your approval, then scaffolds the project and writes the code, and finally builds the application to check for compile errors and gives you instructions to run it.

• Understand requirements: the agent asks concise, targeted questions when key information is missing.
• Propose plan: a short summary of the app type, technologies, main features, and visual approach, for your approval.
• Implement: the agent scaffolds the project with shell commands and writes the code, including placeholder visuals where needed.
• Verify: the agent reviews the work against your request and plan, then builds the app and checks for errors.
• Solicit feedback: the agent explains how to start the app and asks for your feedback.

Install the new-app skill

Download the ZIP for the skill from this page. The skill is distributed as a folder that contains SKILL.md. Unzip it into one of two locations, depending on how you want to use it.

For a personal setup available across projects, unzip the skill into ~/.claude/skills/. To limit it to a single project, unzip it into that project's .claude/skills/ folder.

If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

• Personal (all projects): ~/.claude/skills/new-app/
• Single project: .claude/skills/new-app/ inside your project folder
• Claude apps: Settings → Capabilities → Skills, then upload the skill

How do you trigger the skill with a natural request?

You do not need a special command. Describe the thing you want to create in plain language, and the agent will load the skill when your request is about creating something new. The skill's trigger describes requests to create a new application, project, website, game, mobile app, CLI tool, or library from scratch.

Include what you know: the type of app, the main features, and any tools or languages you want. Anything you leave out, the agent may ask you about before planning.

• "Create a new website for a recipe collection with search and a favourites list."
• "Build a small 2D platform game in the browser with placeholder graphics."
• "Make a command-line tool that renames photos by their date taken."
• "Start a new library for parsing CSV files from scratch."

What does a typical session look like?

Expect the agent to pause at two points: when it asks clarifying questions, and when it presents the plan. Answer the questions and reply to the plan with approval or changes. Only after approval does it write the project.

The skill sets default technology choices when you do not name a stack. For example, it prefers React for websites, Python or Go for CLIs, and HTML, CSS and JavaScript with Three.js for 3D games. For mobile apps it names options such as Flutter, Jetpack Compose and SwiftUI depending on whether you need one codebase or native apps.

• Websites: React with Bootstrap CSS and Material Design principles.
• Back-end APIs: Node.js with Express.js, or Python with FastAPI.
• CLIs: Python or Go.
• 2D games: HTML, CSS and JavaScript.
• 3D games: HTML, CSS and JavaScript with Three.js.

What does the skill need to run?

The skill needs the agent to have shell access so it can run scaffolding and build commands such as npm or npx. Its toolset must also include file writing, file editing, shell command execution, and asking the user questions. The skill names these tools as write_file, edit, run_shell_command and ask_user_question.

The skill does not install toolchains itself. You must have whatever your chosen stack requires, such as Node.js, Python, Go, Flutter or Kotlin, already available on your machine.

• Shell access for scaffolding and build commands such as npm and npx.
• Tools: write_file, edit, run_shell_command and ask_user_question.
• Toolchains for your chosen stack (for example Node.js, Python, Go, Flutter or Kotlin), installed by you.

What are the safety notes?

The skill runs scripts, because scaffolding and build steps are run through shell commands. It does not use the network according to its description. Because the agent executes commands in your environment, review the plan and the commands it proposes before approving, and keep the project in a folder you are comfortable with.

Do not use this skill to change an existing codebase. The workflow is designed for greenfield projects, and the agent may overwrite or restructure existing work if the skill is applied to it.

• The skill runs scripts through the agent's shell tool.
• It is documented as not using the network.
• Use it in an empty or new folder, not in a project you have not backed up.

Tips for better results

The more specific your first request, the fewer questions the agent needs to ask. Name the platform, the core features and the visual style you want. If you care about the stack, name it; otherwise accept the defaults in the plan.

Ask for placeholder visuals explicitly if you do not have graphics yet. The skill is designed to use simple shapes or generated placeholders, and to tell you what each placeholder should be replaced with.

• State the platform (web, mobile, desktop, CLI, library, 2D or 3D game).
• Name any technologies you require; otherwise let the defaults apply.
• Read the plan carefully before approving, since implementation follows it.
• After the build, follow the run instructions the agent gives you and report what you see.