QA Agent
Drives a real browser against your live app, and every finding carries evidence.
- Typical run
- ~10 min
- Tools
- Headless browser · HTTP probes
- You provide
- A URL · Crawl depth · Goals
- Approval
- Read-only: nothing leaves the platform
◤ What it does
Drives a real browser against your live app, and every finding carries evidence.
The QA Agent tests a live web app in a real Chromium browser. It maps and crawls the app, logs in with credentials you supply, runs smoke and negative journeys plus the ones you define, and checks WCAG 2.2 accessibility, Core Web Vitals, and OWASP header and cookie hygiene. The result is a graded scorecard where every finding carries a screenshot, console and network evidence, repro steps, and a stable fingerprint so regressions are tracked run over run.
Limits: Public URLs only; logins only with supplied credentials.
◤ Ask it
In your words. No prompts to engineer.
“Check that signup completes with a fresh email and lands on the dashboard, every form validates inline, and no page has console errors.”
“Audit the checkout flow on mobile: no horizontal overflow, tap targets large enough, and the order completes cleanly.”
“Focus on accessibility and Core Web Vitals on the main pages.”
◤ Missions it runs in
- Audit my software
◤ Questions
About the QA Agent
Is it a real browser or a simulation?
A real headless Chromium. Screenshots, console output, and network traces are captured from the actual session and attached to every finding.
Can it test behind a login?
Yes, with credentials you supply for the run. It never guesses or brute-forces logins.
What formats can I export?
Markdown, JSON, and JUnit XML. Findings can also be filed as GitHub issues, and triage decisions survive across runs.
◤ Works alongside
Agents that share its missions.
Code Review Agent
Reads a GitHub repository and reviews architecture, correctness, security, and tests.
Fix Agent
Turns review and QA findings into verified edits and opens the pull request.
Analyst
Reasons over the evidence earlier steps gathered, as a named specialist, labelling every claim.
Evaluator
Re-fetches sources, recomputes figures, and scores every deliverable before you see it.