UX Review Playbook — definition of done for any UI we ship
This is the quality gate for every UI/dashboard/app surface — Drift by Mar AND personal dashboards (World Cup, sports hub, Home HQ, downloads, etc.). Nothing ships / is called "done" until it has been DRIVEN (not just code-read) and graded against the three goals below.
Canonical, always-current version = the Claude Code skill
ux-review(~/.claude/skills/ux-review/SKILL.md). Run it with/ux-review <flow>. This file is the cross-machine mirror per the playbook-driven rule in 00-START-HERE.
The three goals
- Complete / efficient / coherent flows — finishes what it starts, fewest steps, no dead ends or contradictory states.
- World-class look & feel — formatting, motion, feedback, input ergonomics at best-consumer-app level. No scaffolding.
- Robust under any usage — no input/timing/nav/permission/network condition causes a crash, corruption, silent failure, or misleading state.
"Works on the happy path" is the starting line, not the bar.
How to run it
- Start the REAL app (read its
devscript for the real port). Real browser (Playwright/manual), devtools console + network open the whole drive. Seeded test data realizing every state (empty/1/many, each status, expired/used tokens…). - Two passes per screen: Quick (formatting/clarity, client-side validation, world-class inputs, perceived responsiveness, navigation/orientation) + Robustness (input hardening, concurrency/timing, authorization, console/network hygiene, environment matrix incl. dark mode + 390px + keyboard-only).
- Report findings as
[blocker|rough|nit] <screen> — <what> — <repro> → [fix|accept|defer], close with a per-flow verdict (Goal 1/2/3 ✅⚠️❌) + the robustness checks that held. Evidence or it didn't happen (screenshots / response bodies / console excerpts). - Small obvious fixes inline, then re-drive (unverified until observed). Update the project QA tracker before ending.
Where the gate lives
- Global rule:
C:\Users\marit\CLAUDE.md— UI builds aren't "done" until/ux-reviewpasses (blockers fixed, re-driven). - Per-project (driftbymar): keep a
.planning/ui-qa/tracker (FLOWS.md, QUALITY-RUBRIC.md, DRIVE-CHECKLIST.md, punch list). Project tracker tiers/seeds override the generic skill lists. - Append validated UX learnings to
LEARNINGS.md(Tried / Result / Reuse).