Skip to content
Codex DesignBETA

Frontend Testing Debugging skill for Codex.

Frontend Testing Debugging makes Codex prove a UI change works in the browser, not just in the build. It opens the page, runs a fixed set of checks, exercises the flow you changed, and reports back with screenshots. It ships in OpenAI’s Build Web Apps plugin.

Updated November 12, 2026 · 3 min read

At a glance

Skill namefrontend-testing-debugging
Comes withThe Build Web Apps plugin, by OpenAI
Installcodex plugin add build-web-apps@openai-curated, or /plugins in Codex
Codex uses itWhen you test, debug or improve a rendered frontend: dev servers, UI regressions, broken clicks, console errors, layout
Sourceopenai/plugins on GitHub

What it does

For any change to a rendered page, the skill runs the same loop: name the flow under test in one sentence, pick a browser, make the smallest useful edit, check the result, then report.

It uses the Codex Browser plugin when it’s available. If it isn’t, it falls back to Playwright and says why. Before it claims the app works, it checks six things:

  • the URL and title are the right page
  • the page isn’t blank
  • no Next.js, Vite or Webpack error overlay
  • no relevant errors or warnings in the console
  • a screenshot backs up any visual claim
  • at least one interaction in the flow, followed by a state check

It also looks for clipping, overlap, wrapping, missing assets and z-index problems on desktop and one mobile size. The final reply follows a QA report: summary, environment, a pass/fail table of the checks, the interaction tested, remaining risk, and screenshots grouped at the end. It doesn’t write reports or screenshots into your repo unless you ask.

When to use it

Use it any time Codex changes a UI you can see. It catches the common case where the build passes but the page is blank, throws an error, or the button does nothing.

Try it

Fix and verify a broken interaction
Use the frontend-testing-debugging skill. The search filter on the transactions page of our dashboard doesn't update the table. Find the bug, fix it, and verify it in the browser at desktop and mobile sizes.

Does it work in Claude Code?

Partly. The preferred path uses the Codex Browser plugin and its agent.browser API, which Claude Code doesn’t have. The Playwright fallback is ordinary shell commands, so if you copy the folder into ~/.claude/skills, Claude Code can follow that path and the QA report format.

Keep reading

An independent plugin for Codex. Not affiliated with or endorsed by OpenAI.