One target user, the problem they need solved, and a realistic first journey.
A short brief and a plan of small ordered tasks, each with a visible result and a check, ready for your agent.
Every in-scope behavior has a screen, data requirement where needed, and an observable acceptance check.
Planned: coming to the library.
A loose request leaves the agent to guess, and a "looks good" gives you nothing to check against. This skill turns your brief into small implementation slices with observable acceptance checks before any code is written.
You define one useful journey, its data and its checks. The agent inspects your project, its instructions, current changes and run commands, then lists screens, stored fields, empty and error states, deferred scope and acceptance IDs, and names the assumptions that would change the first version. During planning it changes no application files.
It guides the decisions; it does not choose a stack for you. Each slice has a visible result and a check, and you build one at a time, comparing the result with the brief.
You need a coding agent with access to your project files. Provider accounts, model weights, GPU compute and hosting are not included.
No reviews yet.
Sign in to rate this Vibe.
Learn the loop of brief, edit and review on one small change, with Claude Code launched in the right folder and the result checked.