← Skills

Have your coding agent audit your UX

2026-07-02

You just shipped a feature. Does the onboarding flow actually work? Does the error state make sense? Does the layout hold up on mobile? The honest way to find out is to click through it yourself, step by step, and look at what's actually on screen — not just check that the tests pass.

That's slow and easy to skip. It's also exactly the kind of repetitive, observational work a coding agent is good at, if it can see the screen.

What this skill does

This skill turns your coding agent into a QA walkthrough partner. Point it at a URL and a journey ("test the signup flow", "check the checkout page works on mobile"), and it will:

  1. Navigate the journey step by step — landing page, login, the core action, error paths, and scrolled views (mobile too, if you can narrow the viewport or test on a real device).
  2. Take a screenshot at every meaningful step and actually look at it (not just check the DOM), scoring it against a fixed rubric: clarity, feedback, progress, error visibility, navigation, performance feel, empty states, mobile.
  3. Produce a structured report: a journey map of expected-vs-actual, a critical/moderate issues list with the screenshot and impact for each, what's already working, and — this is the part that matters — every fix found, in priority order, with the file and line where known. Not just the top three.

The result reads like a QA report a senior designer would write after using the feature themselves, except it took your agent a few minutes instead of your afternoon.

…have your coding agent do it, on ThinkRun

The catch with agent-driven UX auditing is that most agents can't actually see anything — they read code, not screens. This skill runs on ThinkRun, which gives your agent a real browser to drive: it can navigate, click, type, and — critically — take a screenshot and read it back as an image, so it's judging what a person would actually see, not inferring from markup.

Two things that matter here:

  • It can test flows that need your real login. ThinkRun's local mode drives your own already-authenticated browser tab, so your agent can audit an app behind a login without you handing over credentials — and it never types your password for you; sign-in stays a human step.
  • Every audit is a recording, not a guess. Because it runs on ThinkRun, the walkthrough is captured as a replayable trace — if the report says step 4 is broken, you (or your agent, later) can go back and see exactly what the screen looked like at that step.

For public-facing pages with no login involved, the same skill runs in ThinkRun's cloud mode instead — an isolated browser session with no local setup required.

One skill, a bigger idea

A UX audit is what a real, logged-in, recorded browser makes possible: your agent doing the same "click through it and look" work a person would do, but on demand, every time you ship. The same foundation — real browser, real screenshots, a full recording — is what the rest of the ThinkRun skill library builds on.

Install

Download the file and drop it into your agent's skills directory (for Claude Code: .claude/skills/ux-audit/):

So the layout is .claude/skills/ux-audit/SKILL.md. Then just ask your agent to "audit the signup flow", "check if the checkout page is broken", or "do a UX review of the dashboard."

Prerequisite: this skill drives ThinkRun, so your agent needs it installed first — the Chrome extension for local mode (real, logged-in browser), or just npx @thinkrun/mcp / the CLI for cloud mode. Full setup: docs.