← ClaudeAtlas

ux-walkthroughlisted

Drive a real browser through a user workflow, capture per-step UX evidence, then hand it to design-wren for a design-grade UX report. Use when asked: "run a UX walkthrough", "walk through the signup flow", "audit this journey", "UX report on the onboarding flow", "friction audit", "where does this flow break down", "evaluate the user experience of X", "how does this flow feel", "test the checkout journey end to end", "re-run the <name> walkthrough", "run the saved UX flow". Steps come from a saved flow (`.sector137/ux-flows/`), a natural-language description, or a prototype/journey blueprint; new flows can be saved for reuse as repeatable UX regression checks.
sector137/crew · ★ 0 · Web & Frontend · score 58
Install: claude install-skill sector137/crew
# UX Walkthrough ## Why this exists Most reviews answer "does one screen render correctly?" This skill answers a different question: **"does the whole flow *feel* right?"** It drives the browser through a multi-step workflow the way a user would — click by click, field by field — and records where the experience snags: dead ends, surprise states, slow waits, ambiguous next steps, lost scroll position, error copy that doesn't help. This skill **captures evidence; it does not render the verdict.** The design judgment belongs to Wren. The walkthrough produces a clean, factual evidence bundle and hands it to the design-wren agent, who turns it into a severity-ranked UX report against the project's design principles. Keep the two jobs separate: a captor who also editorializes produces a worse report than a neutral one. --- ## Step 1: Resolve the flow into ordered steps **First, check for a saved flow.** Saved walkthroughs live in `.sector137/ux-flows/*.json` (project root). Glob that directory: - If the user named a flow ("run the **signup** walkthrough", "re-run the checkout audit"), load the matching file and use its steps directly, skipping resolution. - If the user was vague and saved flows exist, list them by name + description and **ask which to run, or whether to define a new one.** Don't guess. - If no saved flows exist (or none match and the user wants a new one), resolve from scratch via source A or B below. A loaded flow is still confirmed before walking (r