← ClaudeAtlas

verifyinglisted

For what the end user sees — a separate agent checks written pages against their checklist and the rest of the project, and reports what differs. Use when UI has been written, before calling it done or opening a PR, or when asked whether pages match ("does this match", "did I miss a page", a page that looks or behaves unlike its neighbours).
kamenkurtev/ui-consistency · ★ 0 · Web & Frontend · score 70
Install: claude install-skill kamenkurtev/ui-consistency
# Verifying against the checklist ## Overview A separate agent answers one question: **which of these pages do not look and behave like the reference and the rest of the project.** It reports only what differs, and names what it could not check. **Open a linked file, or invoke a named skill, only when a part you are carrying out sends you to it, never before** — whatever the request that handed you this skill says about its links. That part is not done until the file is opened or the skill invoked. ## Steps 1. A separate agent checks, never the author — *Never the author*. 2. It reads the checklist and the page first, and the project only where they do not answer — *What the checker reads*. 3. Before its result is trusted, the dispatcher — never the checker — proves the check on a copy with one planted difference — *First, prove it can see*. 4. It compares region by region and reports only what differs, judged as *Reading a difference* says — *Compare region by region*. 5. It names what it could not check, and the cost — *Say what was not checked*. 6. After the last page of a kind, it compares all the pages of that kind together — *The whole set, at the end*. ## Never the author - **The agent that wrote a page does not verify it.** The checker is a separate agent: a subagent where the harness has them, otherwise a new session given only what is below. - A fresh turn in the same session is the same agent with the same context, and does not count. - If