← ClaudeAtlas

speclisted

Turns a one-line product idea into a spec, a data model and a prd.json backlog, so setup-project and auto can run without a manual planning hop. Use when the user describes something to build rather than a change to make.
djnsty23/claude-auto-dev · ★ 3 · Testing & QA · score 65
Install: claude install-skill djnsty23/claude-auto-dev
# spec One sentence in, three artifacts out: `SPEC.md`, a schema, and a `prd.json` backlog `auto` can work. This is the step between "build me X" and a running sprint, and it exists because `setup-project` deliberately refuses to invent one — it will scaffold a project and skip `prd.json` entirely unless a plan already exists. This skill is where the plan comes from. ## The failure mode this skill is designed against A planning skill does not crash. It emits confident filler: *Auth flow*, *Dashboard layout*, *Set up the database*, *Polish UI*. That list looks like a plan, fits any product ever conceived, and its cost lands later — `auto` grinds through stories nobody can tell are finished, and the sprint ends with everything green and nothing working. So the bar is not "produced a backlog". It is **every story names something a person can do, and says how you would know it works.** `check-spec-output.js` enforces that mechanically; the rules below are how to satisfy it honestly rather than by wording around it. ## Step 1 — Infer hard, ask at most once The point of this skill is that a sentence is enough. Infer the obvious: "habit tracker with streaks" implies users, habits, daily check-ins, a streak derived from check-ins — do not ask about any of that. Ask only where the answer **changes the data model or the security model**, and ask once, in a single `AskUserQuestion` call with at most two questions: | Ask about | Because | |---|---| | Multi-user vs single-user | D