Start your product

The kit is everything a product needs that is not the product. To turn it into yours, describe what you are building and let your agent plan it on top of what is here, with the start-product skill.

Describe it

Paste your idea, a spec or a list of notes, and say what you want:

Turn this into an uptime monitor. People add the URLs of their sites, we check each one every few minutes, and email them when one goes down and when it comes back. Free for 3 monitors checked every 5 minutes; Pro, $12 a month, for 50 monitors checked every minute.

Claude Code picks the skill by itself. With another agent, add “follow .claude/skills/start-product”.

Answer a few questions

The agent asks only what changes what it builds, each with an answer it suggests:

• Personal accounts or teams (which to choose).
• What each plan includes, and what is counted per month.
• What happens to the kit’s own examples: the assistant, API keys and the audit log.
• Whether the mobile app is part of the first version.

Read the plan

Before any code, it writes PRODUCT.md at the root of your repository:

• Decisions: your answers.
• Pieces: each thing the product needs, and the part of the kit that builds it. Monitors are a resource, checks are a background job, alerts are emails, the paid plan is a gated feature.
• Doesn’t fit the kit: what the kit has no pattern for, with the options. For the uptime monitor: a public status page anyone can see without signing in, and a limit on how many monitors each plan allows. Nothing here is built until you choose.
• Slices: the order it will build in, each one small enough to try.

Change anything you disagree with, then tell it to go. Changing the plan now takes minutes; changing a table after it is built takes much longer.

Try each slice

It builds one slice at a time with the skills, and runs bun run check and the browser journeys for each one. After each slice it says what you can try. Review the change as you would a colleague’s, commit it, and let it go on.

PRODUCT.md stays in the repository. Your agent reads it before working on the product, and the same skill adds a large new area later: “Add incidents to the product”.

What is left for you

The agent finishes with a list of what only you can do: secrets in .env, your prices in the payment provider (bun run billing:sync), screenshots of your product for the marketing site, and the words about your product that only you can write.