Skip to main content

Project: QA Browser Agent

  • Difficulty: Advanced
  • Primary stack: TypeScript backend with Playwright
  • Estimated duration: 2 weeks
  • Primary hiring signal: agentic QA automation
  • Primary monetization signal: internal QA copilot starter

Problem statement​

Many teams want AI agents to validate front-end quality, but they jump straight into risky browser automation without clear safety rules. This project teaches a safer path: explicit scenario contracts, allowed targets, dry-run behavior, and inspectable evidence.

Project implementation​

Project implementationP06 — QA Browser Agent
Runnable

Stack

TypeScriptExpressPlaywrightZod

Folder

Open project folder

What it proves

You can build safe browser automation for agents with dry-run policy, evidence capture, reports, and deterministic local validation.

Next upgrade path

Add a richer operator view in Project 05 and orchestrate QA runs from Project 03.

Key commands

npm ci
npm run typecheck
npm run build
npm run smoke
npm run eval
npm run browser-smoke

Core workflows​

  • accept a browser QA request
  • apply environment and safety policy
  • run or simulate a Playwright scenario
  • capture screenshot and report JSON
  • expose run status to other systems

Milestones​

  1. request and safety model
  2. Playwright runner and local evidence capture
  3. reports, traces, and deterministic evals
  4. orchestration and UI integration

Acceptance criteria​

  • safe playground runs succeed locally
  • prod requests are dry-run only or blocked
  • invalid targets are rejected
  • screenshots and reports are inspectable after each run

Starter implementation​

Starter code is now available in projects/p06-qa-browser-agent. The current starter provides a local Express service, Playwright runner, strict safety model, mock-safe scenarios, local reports, screenshots, and a minimal evaluation loop.

The current version is intentionally constrained:

  • it does not use real credentials
  • it does not hit real production systems
  • it blocks destructive requests in v1
  • it uses dry-run mode and mock:// targets to keep CI deterministic
  • CI stays dry-run-safe by default, while an optional browser-smoke script is available for local Playwright validation after installing Chromium

Portfolio packaging​

Show one dry-run flow, one safe executed flow, a report JSON, and a screenshot artifact. The strongest demos pair this backend with Project 05 so the QA run is visible in a copilot-style UI.

Monetization path​

This can become an internal QA automation starter, a browser validation service for agent platforms, or a premium template for teams building safe operator-facing QA agents.