Skip to content

Guide · Agent skill

How to use the Throwaway Prototype Builder agent skill

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

The Throwaway Prototype Builder is an agent skill, a set of instructions that an AI coding agent loads when a task matches it. When you have a design question, this skill tells the agent to build a short-lived prototype to answer it instead of writing production code. This guide explains what the skill does, how to install it, how to trigger it, and where its limits are, so you can use it on your first day.

Throwaway Prototype Buildermattpocock/skills · MIT

Quick start

  1. 1Download the ZIP. Get the prototype skill ZIP from this page. It contains SKILL.md, LOGIC.md, UI.md and agents/openai.yaml.
  2. 2Unzip into a skills folder. Unzip into ~/.claude/skills/ for all projects, or into .claude/skills/ inside one project.
  3. 3Or upload in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the skill.
  4. 4Describe your design question. Ask in terms of whether logic or a state model feels right, or what a UI should look like.
  5. 5Open or run the prototype. Double-click the logic HTML file, or run the UI prototype with the project's task runner command.
  6. 6Fold the decision into real code. Implement the validated decision in the main codebase and keep the prototype on a throwaway branch with a pointer from the issue.

What the prototype skill makes the agent do

The skill is for design questions. It has two branches, and the agent picks one based on the question. If you are asking whether logic or a state model feels right, the agent builds a single shareable HTML file with free-play buttons and tabbed guided walkthroughs that push the state machine through hard-to-reason-about cases. If you are asking what a UI should look like, the agent generates several radically different variations on a single route, switchable through a URL search param and a floating bottom bar.

  • Logic branch: one HTML file you can open by double-clicking, with no install needed.
  • UI branch: several variants on one route, started with a command from the project's task runner.
  • Every prototype is marked as a prototype and keeps state in memory by default.
  • After every action or variant switch, the full relevant state is shown so you can see what changed.
  • When the question is settled, the validated decision is folded into the real code and the prototype is kept on a throwaway branch out of main.

Install the prototype skill

Download the ZIP for the skill from this page. The skill is a folder containing SKILL.md, LOGIC.md, UI.md and agents/openai.yaml. You can install it in one of two ways, depending on where you want it available.

For a personal install across projects, unzip the download into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder. You can also upload the skill in the Claude apps under Settings → Capabilities → Skills.

  • Personal: unzip into ~/.claude/skills/
  • Single project: unzip into .claude/skills/ inside the project
  • Claude apps: Settings → Capabilities → Skills, then upload

Trigger the skill with a natural request

You do not need a special command. The skill describes itself as useful when you want to sanity-check whether a state model or logic feels right, or explore what a UI should look like, so ask in those terms. Describe the question, not the code. If the question is ambiguous and the agent cannot ask you, it will choose the branch that matches the surrounding code (a backend module leads to the logic branch, a page or component to the UI branch) and state that assumption at the top of the prototype.

  • "Does this order state machine handle a refund arriving after a cancellation? Build a prototype so I can click through it."
  • "Show me a few very different layouts for the settings page before we commit to one."
  • "I'm not sure this retry logic feels right. Prototype it so a designer can play with it."

Examples of what you get back

For a logic question, expect a single HTML file that a non-developer can drive. It uses plain language for buttons and walkthrough tabs, not the code's internal names. For a UI question, expect a route in the project's existing routing convention, with several variants you switch between from a floating bottom bar. The skill prefers putting the UI prototype next to the real app so the header, sidebar and data look realistic, because a throwaway route on its own makes every variant look fine.

What the skill needs and what it does on your machine

The source lists no required prerequisites. The skill runs scripts and does not use the network. The UI branch starts from one command in the project's task runner, such as pnpm followed by the prototype name, and the logic branch is a single HTML file you open in a browser. Prototypes keep state in memory. If a question involves a database, the skill is meant to use a scratch database or local file with a clear name such as one that says it should be wiped.

  • Runs scripts: yes. Review the agent's actions before accepting them.
  • Uses network: no.
  • Throwaway code may be added and committed to a separate branch, so check what the agent creates in a project where routes or branches need review.

Safety notes and limits

Prototypes are deliberately rough. The skill skips tests, error handling beyond what makes a prototype runnable, and abstractions. Do not ship a prototype. Also, the skill is a poor fit for production features that need tests, error handling or persistence. Use it to learn something quickly, then write the real implementation through your normal process.

  • Skips polish, tests and persistence on purpose.
  • Do not use it for production features.
  • Review any new route or branch it creates if your project requires review first.

What people use it for

  • Checking a tricky order or subscription state machine for edge cases before writing it.
  • Letting a designer or product manager press buttons through a workflow in a single HTML file.
  • Comparing several radically different dashboard or settings layouts inside the real app shell.
  • Recording the verdict of a design experiment in an issue so the reasoning survives after the prototype is discarded.

Frequently asked questions

How do I install the prototype 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. You can also upload it in the Claude apps under Settings → Capabilities → Skills.

Does the prototype skill require anything else to be installed?

The source lists no required prerequisites. The logic prototype is a single HTML file you open in a browser, and the UI prototype starts from a command in the project's task runner.

Is the prototype skill free to use?

The skill is published under the MIT licence. The source does not mention pricing, so check the licence and the page for terms before reuse.

Does the skill use the internet?

No. The skill is listed as not using the network, though it does run scripts, so review what the agent executes.

Can I use this skill to build a production feature?

No. It is designed for throwaway experiments that skip tests, error handling and persistence. Implement the validated decision in the real codebase using your normal process.

What is the difference between the logic and UI branches?

The logic branch builds a single HTML file for testing state transitions by clicking buttons. The UI branch builds several radically different layout variants on one route, switchable from a floating bottom bar.

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 when you have a question about whether logic feels right or how a UI should look, and you want fast answers without writing production code. Skip it for production features that need tests, error handling or persistence, and for projects that do not want an agent creating throwaway routes or branches without review.

Download the skill