audit

audit

Audit the WiM onboarding implementation against the encoded brand and architectural rules.

Trigger phrases

• "audit the onboarding"
• "what's wrong with the onboarding flow"
• "check onboarding for brand drift"
• "review the existing onboarding screens"
• "find issues in the welcome flow"

What to check

Files in scope

• app/src/main/assets/onboarding.html
• app/src/main/assets/onboarding.js (if present)
• app/src/main/assets/onboarding.css (if present)
• app/src/main/java/com/wim/app/**/onboarding/**/*.kt
• app/src/main/res/values/colors.xml and app/src/main/res/values-night/colors.xml
• app/src/main/res/raw/mrdna_*.json (Lottie assets)
• app/src/main/res/font/* (Limelight presence)

Drift checks (in priority order)

P0 — brand integrity

1. Limelight on every header. Grep for headers that use Outfit/Roboto/system default. List them with file:line.
2. "Stutter" anywhere in user-facing copy. Grep for the word in strings.xml, onboarding.html, *.kt. List file:line.
3. Color values match brand.local.md exactly. Grep for hex values not in the canonical list. Common drift: #FFD4C8B0 written as #D4C8B0 (alpha dropped) or #D5C9B1 (rounded).
4. Mr. DNA persona present where expected. Onboarding screens that mention WiM by name without the bubble are usually missing the persona.

P1 — animation correctness 5. tween() easing applied to anything tagged Mr. DNA. Grep tween in onboarding/**/*.kt and flag every match for review. 6. Lottie compositions referenced but the JSON file does not exist in res/raw/ or assets/. Glob mrdna_*.json and cross-check with LottieCompositionSpec.RawRes(R.raw.mrdna_*) references. 7. Animations on layout properties (width, height, top, left) instead of transform/opacity. CSS-only check; warn with file:line.

P2 — architectural correctness 8. Permission status checks missing. The onboarding flow must check Settings.canDrawOverlays, accessibility-enabled, mic permission, battery-optimization-ignored, and Samsung-specific status on resume. If any is missing, the "Done" state will never light up. 9. BubbleService not invoked from interactive demo. Demo screens that mock the response are wrong — flag them. 10. Hardcoded strings instead of R.string.*. Cheap to skip, expensive when localization gets added later.

P3 — quality polish 11. Duplicated CSS rules in onboarding.html. Common after iterating tile-by-tile. 12. Unused CSS classes. Grep declared .foo against used class="foo". 13. Tile count in .slider { width: Xvw } mismatched with actual <div class="tile"> count.

Output format

Report as prioritized punch list:

P0 — brand integrity
  ✗ onboarding.html:142 — heading uses font-family: Outfit, should be Limelight
  ✗ OnboardingActivity.kt:88 — string "Help with stutter" should be "speech disfluency"
  ✗ colors.xml:14 — copper_gold is #FF8B5A2F, brand.local.md says #FF8A5A2F (1 hex off)
  ✓ Mr. DNA present on tile 1, 3, 4

P1 — animation
  ✗ MrDnaBubble.kt:45 — tween(durationMillis=300) used; Mr. DNA must use spring()
  ⚠ onboarding.html:201 — animating top property; switch to transform: translateY

P2 — architecture
  ✗ InteractiveDemoScreen.kt:120 — fakes ASR result with hardcoded string; must call BubbleService
  ✓ Permission status checks present for all 5 cards

P3 — polish
  ⚠ onboarding.html — 3 unused CSS classes (.legacy-card, .v1-tile, .old-cta)

Use ✗ for must-fix, ⚠ for should-fix, ✓ for "checked, clean."

Do not produce a wall of text. Do not write a narrative report. The user reads punch lists; they don't read essays.

How to run

For each check, use Grep / Glob / Read directly. Do not delegate to a sub-agent — the audit IS the agent's role; nesting another agent for a Grep is waste.

If the user wants the issues fixed, ask once: "Fix the P0 items now?" — if yes, fix them and re-run the audit. Do not auto-fix without asking.

Quality bar

• All P0 items reported with file:line
• No false positives — verify each finding by reading the actual line, not just the grep match
• Re-runs are quick — cache nothing