DevelopmentIntermediate

From idea to a spec an AI can actually build

Turns a vague idea into scope, data model and acceptance criteria before any code is written.

Works well with

When to use it

Right before opening Cursor, v0 or Claude Code. Skipping this step is the number one reason projects never finish.

Copy Prompt

Copy the prompt and replace the placeholders in braces.

Act as the tech lead who will receive this idea and build it with a coding agent.

Idea: {{idea}}
Main user: {{user}}
Constraints: {{constraints}}

Do not write code yet. Deliver:

1. PROBLEM in one sentence and the work the user does by hand today.
2. MVP: 5 features maximum. Mark what is out of v1 and why.
3. Main FLOW step by step, from the moment the user enters until they get value.
4. DATA MODEL: entities, fields, types and relationships. In text or SQL.
5. SCREENS: a list with the purpose and key elements of each one.
6. Numbered ACCEPTANCE CRITERIA, verifiable without interpretation (e.g. saving a record without an email shows an error and does not submit).
7. Technical RISKS and the default decision for each one.
8. BUILD ORDER in 5 milestones, each one deployable.

Rules: if something is ambiguous, ask me questions before assuming. It is forbidden to add features I did not ask for. Use the {{constraints}} stack if one is defined.

Variables to fill in

Variable	What it means	Example
`idea`	The idea in your own words	An app for photographers to send private galleries to clients
`user`	Who uses it	Independent wedding photographers
`constraints`	Stack, budget or limits	Next.js, Supabase, free, 2 weeks

Tips

Save the spec as SPEC.md at the root of the repo and give it to the agent in every session.
Acceptance criteria are what stop the AI from saying "done" when it does not work.
If the model proposes 12 features, cut it down: the 5-feature MVP is the one that gets finished.

Editorial template. Test it with real data before using it in production.