Integration

AI user testing in GitHub Actions

One workflow file. Every preview deploy gets used by AI agents before review, and the pull request gets one comment that stays up to date.

5 agents try to .

First test freeRun it from your coding agent

Is there a GitHub Action for AI user testing?

Yes: the Mimiq preview check (github.com/victorgulchenko/mimiq-preview-check). It runs on deployment_status events from Vercel, Netlify and Cloudflare Pages, or on any URL you give it. AI agents use the preview in real browsers, and the Action keeps one pull request comment with how many finished, where the rest got stuck, what the browser recorded, and a link to the full report. It can fail the check when agents give up.

Checks that never look at the app

Most pull request checks lint, type-check and run unit tests. None of them opens the app. The first time anyone uses the change as a user is after merge, sometimes in production.

Composite action, no dependencies.
Inputs: url, goal, audience, people, max-steps, fail-on.
Access: vercel-bypass-secret, basic auth, login-email and login-password.
Outputs: report-url, finished, partway, gave-up.
Without a key, a small free check of up to 3 agents runs.

How it works

01
Add the secret
MIMIQ_API_KEY from Settings, "Create a key".
02
Add the workflow file
.github/workflows/mimiq.yml, below.
03
Tune it
Set goal to your most important task, people to 3 or 5, and fail-on once you trust it.

The workflow

Minimal setup:

on: deployment_status
permissions:
  pull-requests: write
jobs:
  mimiq:
    if: github.event.deployment_status.state == 'success'
    runs-on: ubuntu-latest
    steps:
      - uses: victorgulchenko/mimiq-preview-check@v1
        with:
          api-key: ${{ secrets.MIMIQ_API_KEY }}
          goal: Sign up and reach the dashboard
          people: 3

Testing a fixed URL instead

Pass url to test staging or production on a schedule (on: schedule) or on push, without a deployment event.

Limits

Agents are simulated users. They find problems fast and on every build; they do not forecast how many real people will convert, they are not a deterministic regression suite, and they do not install native mobile apps.

40steps at most for each agent, each in its own real browser. Every step comes with a screenshot and what the agent was thinking at the time.

20agents at once, each acting as a different one of your users. The browser also records every error, failed request and request slower than 3 seconds on your site, on the step it happened. Agents find problems; they do not forecast how many people will convert.

$0for your first test, with a free account (Google sign-in, no card). It includes 50 credits, enough for 10 agents. After that an agent uses 5 credits, and packs start at $9 for 120 credits.

See a sample report

Questions

Who posts the comment?

The workflow's GITHUB_TOKEN, with pull-requests: write. One comment, found by a marker and updated on each push.

What if I run out of credits?

The Action posts a comment that says so, with a link to add credits, and does not fail your build unless you asked it to.

Keep reading

Let the agents find what breaks first.

Your first test is free: sign in with Google and it starts right away. Last checked 2026-10-22.