📱 Mobile UX Audit

An agent-agnostic UX audit toolkit for mobile apps. Point it at a mobile codebase (or screenshots / Figma designs) and it produces a prioritized, evidence-backed UX audit — and, on request, fixes the issues.

Supports React Native / Expo, Flutter, SwiftUI / UIKit, Jetpack Compose / XML, Kotlin Multiplatform, .NET MAUI / Xamarin, Ionic / Capacitor / Cordova, and PWAs — with webview-specific checks for the last two.

Design principle: the audit logic lives in one portable Markdown "brain" (playbook/ux-audit-playbook.md). Tool-specific wrappers (SKILL.md, AGENTS.md) are thin and just tell each tool when to load the brain. So this works with Claude today and any other agent tomorrow — with zero changes to the logic.

What it checks

Touch targets & interaction · loading/empty/error states · accessibility (labels, contrast, dynamic type) · visual hierarchy & spacing · safe areas & responsiveness · forms & input · navigation & flow · feedback & motion · content & copy · iOS HIG / Android Material platform fit.

Each finding gets a severity (P0–P3), a confidence tag (verified / likely / needs-runtime, with a how-to-verify hint), an effort estimate, an evidence pointer (file:line), and a concrete fix. See references/heuristics.md and a full example report.

Built for real dev workflows

• Batch mode — large apps are audited in screen batches with findings flushed to a disk ledger, so token cost stays flat as the app grows (playbook §6).
• Structure trace — optionally reads the stack's routing config (and GraphQL schema if present) to order batches by real user flow and catch orphan/dead-end screens (Step 2.5).
• Re-audit diffing — run it again next sprint; the report leads with Δ since last audit: 4 fixed · 2 new · 1 regressed (playbook §6.5).
• Reviewable fixes — --fix lands one commit/diff per finding, never a merged blob.
• Ticket export — --tickets emits paste-ready GitHub/Jira issue blocks for every P0/P1.

Repo layout

mobile-ux-audit/
├── SKILL.md                        # Claude Code skill wrapper (/audit-ux)
├── AGENTS.md                       # Wrapper for Cursor / Copilot / other agents
├── cli/                            # standalone CLI (audit + screenshot-diff), Claude API
├── ux-audit.config.example.json    # per-repo config: categories, failOn gate, ignore globs
├── playbook/
│   └── ux-audit-playbook.md        # ← the portable brain (the real logic)
├── references/
│   ├── stack-detection.md          # detect framework + fix cheatsheet + verify hints
│   ├── heuristics.md               # Nielsen + mobile principles + a11y + anti-patterns
│   └── platform-guidelines.md      # iOS HIG vs Android Material
├── templates/
│   └── audit-report-template.md    # the report format (findings, fixes, tickets)
└── examples/
    └── example-report.md           # what the output looks like

Use it with Claude Code (published skill)

Install as a skill, then run /audit-ux in any mobile project.

Global (all your projects):

git clone https://github.com/aditya305/mobile-ux-audit ~/.claude/skills/audit-ux

Or per-project:

git clone https://github.com/aditya305/mobile-ux-audit .claude/skills/audit-ux

Then in Claude Code:

/audit-ux                       # report only
/audit-ux --fix                 # audit, then apply fixes (one commit/diff per finding)
/audit-ux --batch               # force batch mode (auto for ~10+ screens)
/audit-ux --trace               # trace nav/GraphQL structure first, order by user flow
/audit-ux --tickets             # also emit GitHub/Jira issue blocks for P0/P1
/audit-ux --screens login home  # limit scope

Run it again next sprint against the same ledger and the report opens with what got fixed, what's new, and what regressed.

Use it as a standalone CLI (CI-ready)

No agent needed — the CLI calls the Claude API directly with the same playbook, using prompt caching so the playbook is billed once and read from cache on every batch.

npm install && npm link          # or: npx mobile-ux-audit (once published)
export ANTHROPIC_API_KEY=...     # or `ant auth login`

mobile-ux-audit ./my-app                    # audit → ledger + report + exit code
mobile-ux-audit ./my-app --dry-run          # inventory + batch plan, no API calls
mobile-ux-audit ./my-app --fail-on P1       # CI gate: exit 1 on P0/P1 findings
mobile-ux-audit diff --baseline shots/v1 --current shots/v2 --analyze
                                            # screenshot-diff regression mode

• Batched with prompt caching — flat peak cost regardless of app size.
• Re-audit diffing — reruns lead with Δ: fixed / new / still open.
• ux-audit.config.json — tune model, categories, severity overrides, ignore globs, batch size, and the failOn gate (see ux-audit.config.example.json).
• Screenshot-diff mode — pixelmatch baseline vs current; --analyze sends changed pairs to Claude vision for a regression verdict; --fail-on-regression gates CI.

Use it with any other agent

• Cursor / Copilot / Windsurf: drop this repo in your project (or copy AGENTS.md + playbook/ + references/). The agent reads AGENTS.md and follows the playbook.
• Any LLM / manual: attach or paste playbook/ux-audit-playbook.md and say "audit the UX of this app following this playbook." That's it — the playbook is self-contained.

Roadmap

• [x] Batch mode with disk ledger (flat token cost on large apps).
• [x] Structure/GraphQL trace to order audits by real user flow.
• [x] Re-audit diffing (fixed / new / regressed).
• [x] Ticket export + reviewable per-finding fixes.
• [x] Extended stacks: KMP, MAUI/Xamarin, Ionic/Capacitor/Cordova, PWA.
• [x] Standalone CLI (mobile-ux-audit) calling the Claude API, with prompt caching + CI exit codes.
• [x] Screenshot-diff regression mode (pixelmatch + optional Claude vision verdicts).
• [x] Config file (ux-audit.config.json) for severity overrides and enabled categories.
• [ ] Publish to npm (npx mobile-ux-audit).
• [ ] GitHub Action wrapper.

License

MIT — see LICENSE.