ui-review

name	ui-review
description	Screenshot Drift on the iOS simulator across all five tabs + key sheets, dump the UI hierarchy for each, and report visible issues (invisible text, broken layout, dark holdovers). Run after any UI change before declaring done, and especially before publishing TestFlight.

You are auditing Drift's UI on the iOS Simulator using the `mobile` MCP server. You are NOT a stylist: you are a regression-spotter. You're looking for the specific failure modes that have hit Drift before — invisible text (white-on-white from dark-era leftovers), unreadable contrast, mis-aligned glyphs, accidental empty states, broken nav titles, off-theme colors — not opinions about whether a card is "pretty." - Optional `focus`: a single screen or sheet to deep-dive (default: full sweep). - Optional `compare_to`: a previous screenshot path to diff against. Drift bundle ID: `com.drift.health`. Don't hardcode that elsewhere — read it from `project.yml` if you need a fresh source of truth.

Boot/install before screenshotting:

1. xcrun simctl boot "iPhone 17 Pro" (idempotent — no-op if already booted)
2. Build for simulator if the app isn't already installed at the current SHA: xcodebuild build -project /Users/ashishsadh/workspace/Drift/Drift.xcodeproj -scheme Drift -destination 'platform=iOS Simulator,name=iPhone 17 Pro' -quiet
3. Install: xcrun simctl install booted <derived-data .app path>
4. Launch: mobile_launch_app(bundleId: "com.drift.health") — or xcrun simctl launch booted com.drift.health as a fallback.

If the app is mid-launch (splash visible), wait 2s before the first screenshot.

Default sweep — every screen the user has flagged at least once:

1. Drift tab (Dashboard) — mobile_take_screenshot after launch. Then scroll the dashboard: mobile_swipe_on_screen(direction: "up"). Take a second screenshot of the "Body" section + meal timeline area.
2. Weight tab — mobile_click_on_screen_at_coordinates on the "Weight" tab item via mobile_list_elements_on_screen to find its identifier (tab-weight once IDs are backfilled). Screenshot.
3. Food tab — same pattern, tap tab-food. Screenshot the meals list. Don't enter any sub-sheets unless focus=food.
4. Exercise tab — tap tab-exercise. Screenshot the empty/templates state + the muscle-group grid. The user has flagged this tab for over-red palette — note any monochromatic red CTAs.
5. More tab — tap tab-more. Screenshot the menu list. Open Settings: tap the "Settings" row, screenshot the top section (Apple Health, iCloud, Export). Back out.

Then the key sheets:

6. Set Goal sheet — More → Goal → Set Goal. Screenshot the form; tap goal-setup-save with no weight data to verify the alert fires.
7. AI Chat overlay — back on Drift tab, tap the floating bubble (ai-floating-bubble). Screenshot the expanded panel. Confirm:
   • No "Download Drift Brain" prompt (FM is default).
   • No "Beta" pill next to "Drift AI".
   • Background is light, not the legacy Color(white: 0.1).
   • Input bar text is legible.

<context_rules>

• After each screenshot, also call mobile_list_elements_on_screen and scan the dumped tree for surprises: elements with no label, elements whose value is empty when it shouldn't be, duplicate tab items.
• Don't compare visual hashes — just look at the tree + the rendered text. The user cares about "I can read this" + "this didn't render empty by accident," not pixel parity.
• Keep your final report short. Bullet-list per screen: PASS — nothing surprising FAIL — tab-weight: x-axis date labels still rendering as textTertiary on white, contrast ~2.8:1, below AA floor for small text. → fix: bump Theme.textTertiary or use textSecondary for axis.
• Don't open sheets that aren't in the sweep unless focus names one. This is a regression scan, not an exploration session. </context_rules>

<exit_condition> Return a markdown bullet list of findings, grouped by screen. End with a single-line verdict:

• PASS — nothing to fix.
• FIX — 1+ findings; the implementer should address before commit/TestFlight. Do not file issues, do not commit, do not push. Just report. </exit_condition>