Generate your specs
Every MVP feature specified at once, in the right order.
All your specs, written in parallel
Writing feature specs one at a time is slow and you lose context between them. Feature three depends on feature two, which depends on a database table from feature one. If you spec them independently, you get inconsistencies and gaps.
Inside Claude Code, type /mvp-spec. It reads your full discovery output and generates specs for every MVP feature at once. It knows the dependencies, the build order, and which features can be worked on in parallel.
How it works
The system reads your feature map and extracts everything marked as MVP. It checks what's already built (the skeleton features like auth and billing are there from the start) and skips anything that already has a spec.
It determines the build order. Foundation features go first: design system, landing page, auth. Then features that depend on those. Independent features get grouped for parallel generation. The system writes up to three specs at a time.
Each spec is written by an agent that has full context. It reads your product overview, user journeys, UX patterns, brand guidelines, and technical requirements. If the feature uses an external tool, it reads those docs too. The spec includes the user experience (written like a storyboard), business rules, edge cases, test flows, and UI notes.
What you review
When all specs are ready, you get a table showing every feature with its size, dependencies, and build order. A visual sequence shows what builds when and which features run in parallel.
You can walk through each spec one by one, edit specific ones, reorder the build sequence, or drop features to V2. Nothing gets built until you approve the plan.
What comes out
A roadmap.json file that tracks build progress, and a set of ordered spec files that /mvp-build reads from. Each spec is self-contained. An agent reading just that file plus the referenced docs can build the feature without additional context.