E2E testing that actually uses your app.
One command. Your AI agent reads your codebase, opens your real browser, clicks through critical flows, and reports exactly what broke. You get screenshots and actionable findings, not vague test logs.
$ npx hanzi-browse setup $ /e2e-tester
$ npx hanzi-browse setup $ hanzi-browser start "test my checkout and signup flows" --skill e2e-tester
You define the flow. It runs the investigation.
Scans your recent changes
Before opening the browser, it checks recent diffs and routes so testing focuses on what is most likely to break.
Executes real user flows
It opens your app in a real browser, fills forms, clicks buttons, navigates pages, and behaves like a QA tester.
Captures proof at each step
After each flow, it saves screenshots and logs what passed, failed, or looked suspicious so you can reproduce issues quickly.
Correlates failures with code
When something fails, it points to likely files and root-cause clues from your codebase, not just UI symptoms.
Produces a QA report
You get a concise pass/fail summary with severity and next actions, so fixes can be prioritized before release.
Respects safety boundaries
For production or unknown targets, it defaults to read-only checks until you explicitly approve state-changing actions.
Not just a test runner script.
Code-aware + browser-aware
It inspects diffs, routes, and files first, then validates behavior in the real UI. That means faster diagnosis and better bug reports.
Evidence, not guesswork
Each finding includes what happened, where it happened, and screenshot proof. Easy to verify, easy to hand off.
Works with your existing stack
Claude Code, Codex, Cursor, and other MCP clients can run it. No custom QA framework required to get started.
One command. Start testing.
npx hanzi-browse setup