Test Ux

/test-ux — portable UX testing runner

Execute the UX Testing Playbook (playbook.md, same folder) against the current product and emit a findings report (and a real-user test plan). Read it first — it's the method; this is the runbook. Works for any web/app product; a project-level test-ux overrides this when present.

Scope: an argument narrows to a flow/screen (/test-ux "signup flow"). No argument = the top tasks from Phase 0.

Phase 0 — Detect & scope

• Find how to launch the app (project README/TESTING.md, dev script in package.json, or ask). Identify supported locales + text direction.
• Read a project TESTING.md / docs/testing/* / the Project appendix in playbook.md and prefer its goals/tasks/personas.
• Define the top 3–5 tasks as user goals (not features) + a one-line success criterion each; pick ≥1 novice persona.

Runbook (see playbook.md for the why)

1. Scope & journeys — goals, personas (≥1 novice), top tasks, environments (every locale + text direction, mobile + desktop).
2. Heuristic evaluation — Nielsen's 10 per key screen; screenshot evidence; heuristic# · screen · issue · severity 0–4 · fix.
3. Task walkthroughs — drive each task step-by-step; at each step apply the 4 learnability questions; log steps-vs-optimal, dead-ends, hesitation; verdict completed/friction/failed.
4. Journey & states — empty/loading/low-data/error/first-run/returning; error recovery; trust & provenance.
5. Content & i18n comprehension — plain language, no jargon, no raw technical identifiers shown to users, honest tone; per locale verify meaning parity + correct direction + no untranslated source text on the rendered page (not just resource files).
6. Real-user test-plan generator — moderated + unmoderated scripts, screener, metrics sheet (task success, time, SEQ, SUS). Save alongside the report.
7. Report — severity-sorted findings + top-3 blockers, top-3 quick wins, and the explicit "validate with real users" list.

Tools to drive when available

• A browser/preview MCP or the project's dev server for live walkthroughs + screenshots; read components / i18n files via the filesystem for the content + heuristic passes.
• If no live browser is available, do the analytic passes (heuristics, content, i18n from source) and mark the task-walkthrough dimension NOT VERIFIED — never report an un-run walkthrough as passed.

Always

• Walk the full flow in every locale, incl. one RTL — the deep, data-dense screens (results, settings, assumptions), not just the hero. Most i18n/content bugs hide there.
• Anti-degraded-pass: a dimension that couldn't run = NOT VERIFIED in the report, with screenshots of the deep screens for a human glance.

Done when

Report written + real-user test plan saved; findings severity-sorted with evidence + heuristic + fix; severity counts + top blockers printed. Any un-run dimension flagged NOT VERIFIED, not passed.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: GalaxyRuler
• Source: GalaxyRuler/Galactic-skills
• License: MIT

Install and usage instructions live in the source repository linked above.