Write a product specification with AI
Agree the scope and acceptance criteria for a feature before a coding agent turns the design into tasks.
When you write a product specification with AI, the first decision may be how many features you are actually asking for. A request for an event booking system can include reserving a seat, collecting payment and handling cancellations. Each has different failure cases and can be tested separately. The spec-driven-development skill starts with a capability map for requests of that size.
For a focused feature, the agent can go straight to the specification. It records boundaries, project conventions, open questions and observable success criteria. You review that document before technical planning begins. If the scope changes, the specification remains the place to record the decision.
What it does
- Separate independently testable features in an oversized request.
- Describe what a customer can do when the work succeeds.
- Record an explicit exclusion such as refunds in a booking prototype.
- Identify a dependency before the team commits to a build order.
- Give reviewers concrete criteria for deciding whether a feature is ready.
Install and use
- Use a supported coding agent and provide access to the project guidance and intended document location.
- Describe the booking flow, identify its users and name one difficult case. Request a specification for review before implementation.
- Resolve open questions, approve the scope and continue with a technical plan in a subsequent step.
FAQ
Source
Made by Addy Osmani. Original skill: spec-driven-development, published on GitHub under the MIT licence.
View on GitHub