content-audit

name	content-audit
description	Audits PRD-specified content counts (screens, copy strings, images, supported locales, push categories, onboarding steps) against what is actually implemented. Identifies what is planned vs. what is built. Read-only.
argument-hint	[no arguments | --strict]
user-invocable	true
allowed-tools	Read, Glob, Grep
model	sonnet

Content Audit

A PRD says "5 onboarding screens"; the codebase has 3. A PRD says "supported locales: en, es, ja"; the i18n bundle has only en.json. This skill quantifies that gap.

Read-only. Produces a printed report.

---

Purpose / When to Run

Run when:

• Pre-release — confirm content matches plan before submitting to stores
• Mid-sprint — confirm a content-heavy feature is on track
• After a copy push — confirm strings landed
• The user types "content audit" or wants a quick "is the content done" check

Distinct from /asset-audit (in the second skill set, which checks assets specifically) — this skill checks counts and presence, not asset compliance.

Inputs

• All design/prd/*.md
• Source code:
  • src/ for RN/JS/TS
  • lib/ for Flutter
  • ios/ and Xcode project for iOS native
  • app/src/main/ for Android native
• Localization bundles
• Asset directories (assets/, Assets.xcassets, app/src/main/res/)

Outputs

• A printed report. No file writes.

---

Phase 1: Inventory Planned Counts

Read every PRD. From each, extract count claims. Common patterns to grep:

• " screens"
• " onboarding steps"
• "supported locales: "
• "push categories: "
• " illustrations"
• " default templates"
• "feature flags: "
• "tabs: "

Build a planned-content table:

| Source PRD | Item | Planned count |
|------------|------|----------------|
| onboarding.md | screens | 5 |
| onboarding.md | locales | en, es, ja |
| paywall.md | tiers | 3 (monthly, annual, lifetime) |
| home.md | tabs | 4 |
| ... | ... | ... |

If a PRD makes vague claims ("a few", "several"), warn but do not error. List them as unknown and recommend tightening the PRD.

---

Phase 2: Inventory Implemented Counts

Use globs and greps appropriate to the framework. The skill must be framework-aware — read .claude/docs/technical-preferences.md first.

React Native / Expo

• Screens: glob src/screens/**/*.{tsx,jsx} or routes in app/ (Expo Router)
• Onboarding screens: count screens under src/screens/onboarding/ or matching pattern
• Locales: list files under src/locales/, src/i18n/, or assets/locales/
• Asset images: count under assets/images/ and src/assets/
• Components: count under src/components/

Flutter

• Screens: count lib/features/**/screens/*.dart or lib/screens/
• Locales: list under lib/l10n/ or assets/translations/
• Assets: list under assets/images/, assets/icons/

iOS native

• Screens: count *View.swift or *ViewController.swift files
• Locales: count *.lproj directories under Resources/
• Asset catalogs: count entries in Assets.xcassets

Android native

• Screens: count *Activity.kt, *Fragment.kt, or Compose screens (@Composable fun ...Screen())
• Locales: count values-* directories under app/src/main/res/
• Drawables: count under res/drawable-*/

For each PRD claim, find the corresponding implemented count.

---

Phase 3: Compare and Classify

For each row:

• MATCH — planned count = implemented count (or planned ≤ implemented for "at least N" cases)
• UNDER — implemented count < planned count
• OVER — implemented count > planned count (often fine, but flag — may be unplanned scope)
• UNKNOWN — cannot determine implemented count (e.g., copy strings without a strings file)
• PLAN MISSING — implementation exists but no PRD claims it (potential undocumented feature)

Severity:

• HIGH — UNDER on a v1-critical item (locales, payment tiers, onboarding screens, account-deletion flow)
• MEDIUM — UNDER on a v1-deferrable item, OVER on anything
• LOW — UNKNOWN
• INFO — PLAN MISSING

---

Phase 4: Report

# Content Audit

PRDs scanned: <N>
Items audited: <N>
Findings: <count by severity>

## HIGH
1. **Locales** — onboarding.md plans `en, es, ja`. Implementation has `en` only.
   Files: src/locales/en.json
   Action: source translations for `es` and `ja`, or update PRD to reflect ship-with-en-only.

2. **Onboarding screens** — onboarding.md plans 5 screens. Implementation has 3 (Welcome, Permissions, Profile).
   Action: implement Goals and FirstAction screens, or update PRD.

## MEDIUM
- Paywall tiers — paywall.md plans 3 (monthly, annual, lifetime). Implementation has 2 (monthly, annual). Consider whether lifetime is in v1 or v1.1.

## LOW (UNKNOWN)
- Copy strings — onboarding.md says "approximately 40 strings". No strings file found, only inline literals. Recommend extracting to enable localization.

## INFO (PLAN MISSING)
- Implemented but unmentioned: 2 screens under src/screens/debug/. Likely dev-only — confirm and document or remove from production builds.

## Summary
- Plan match rate: <N>%
- Under-implemented items: <count>
- Over-implemented items: <count>
- Unverified items: <count>

---

Phase 5: Strict Mode

If --strict was passed, additionally flag:

• Inline string literals (no localization) — every screen using literal English strings instead of translation keys
• Missing accessibility labels on images (in framework-appropriate way: accessibilityLabel, Semantics(label:), contentDescription)
• Missing dark-mode variants (color resources without dark-mode override)

These run as additional MEDIUM findings.

---

Phase 6: Recommendations

## Recommended actions

For each HIGH:
- If the count gap is intentional: update the PRD to reflect reality (`/design-system retrofit ...`).
- If the gap is unintentional: implement the missing items.

For OVER items:
- Decide whether to deprecate the extras, or update the PRD to claim them.

For PLAN MISSING:
- Document via `/reverse-document --prd <area>` or remove from production builds.

---

Edge Cases

• No PRDs: stop. There is nothing to audit. Recommend /design-system to author the first PRD.
• Implementation does not match the framework declared in technical-preferences: warn loudly. Either the framework pin is wrong or there is hybrid code.
• Counts are subjective (e.g., "a few"): list as UNKNOWN, do not guess.

---

Quality Gates

• Every HIGH finding includes the source PRD path and the implementation path that was inspected.
• Counts in the report are reproducible — running the skill twice on the same project produces the same numbers.
• The skill never edits any file.

---

Examples

/content-audit

• Plan: 5 onboarding screens, 3 locales (en/es/ja), 2 paywall tiers
• Impl: 3 onboarding screens, 1 locale (en), 2 paywall tiers
• Findings: 2 HIGH (locales, onboarding screens), 0 MEDIUM
• Verdict: Major content gap — pre-release blocker.

/content-audit --strict

• Adds: 17 screens use inline string literals instead of translation keys
• 4 images missing accessibility labels
• Verdict: HIGH localization risk; remediation needed before international launch.