← back to work
case study · /qa-agent-skill · open source

QA Agent Skill.

A production-grade Playwright QA automation skill stack for AI coding agents: strict POM architecture, durable locators, no-static-wait synchronization, production safety, feature flags, mocking, CI observability, and security-aware artifacts.

skill stack · open source playwright 23 sub-skills typescript · POM audit scripts
↗ github ↗ medium article ↗ linkedin
Editorial cover for the Playwright QA agent skill stack
// 01 · operating model

Durable automation, not throwaway scripts.

entrypoint
One global qa-agent skill
Agents start from a single index, grep first, then load only the sub-skill and references needed for the task.
architecture
Strict POM-first framework
Specs express business intent. Page objects, components, fixtures, config, and utilities own the implementation details.
enforcement
Contracts plus audits
Shell audits catch static waits, forbidden locators, force clicks, focused tests, and production guard bypasses.
// 02 · hard contracts

What the agent is not allowed to ship.

  • No static waits in final code.
  • No networkidle waits in tests.
  • No text, placeholder, label, role-name, XPath, or DOM-chain locators in final framework code.
  • No raw selectors in specs.
  • No destructive production testing by default.
  • No secrets in committed files, traces, HARs, screenshots, videos, or logs.
  • No accepting generated Playwright code until it is hardened into the framework style.
// 03 · what's inside

Twenty-three indexed sub-skills.

core framework
Project init, POM architecture, test authoring, data fixtures, feature flags, and environment guards.
reliability
Locator contract, synchronization waits, flake prevention, fast automation, CI observability, and maintenance governance.
agent safety
Auth credentials, MCP/CLI guidance, codegen hardening, repo research, review audit, and artifact hygiene.
coverage
API testing, mocking, accessibility, visual baselines, mobile emulation, and Playwright feature selection.
template
A TypeScript Playwright POM scaffold with fixtures, env config, page objects, docs, and audit scripts.
philosophy
The skill teaches an agent when fast code is useful and when fast code becomes automation debt.
// 04 · install surface

It ships with a scaffold and audits.

# create a new strict Playwright POM project
bash bin/init-playwright-pom-project.sh /path/to/new-project

# audit an existing repo before calling it production-ready
bash bin/audit-forbidden-patterns.sh /path/to/repo
bash bin/static-wait-audit.sh /path/to/repo
bash bin/locator-audit.sh /path/to/repo
bash bin/prod-guard-audit.sh /path/to/repo
// 05 · launch post

LinkedIn announcement.

open on linkedin →

The launch post frames the core idea: most automation fails when it treats QA as script generation. This repo turns Playwright work into a governed skill system an AI agent can follow, audit, and improve.

For recruiters and engineering teams, this is the proof point I want visible: I do not just write tests. I encode QA engineering judgment into reusable systems.

linkedin · announcement QA Automation Playwright AI Agents
I built a production-grade Playwright QA agent skill stack.
The post introduces the public repo and the idea behind it: AI agents should not only generate Playwright tests, they should follow QA engineering contracts for locators, waits, production safety, mocks, CI artifacts, and security hygiene.
linkedin.com/in/mrnewdelhi →
// 06 · related

Adjacent work.

AIO Test Case Creator
Another skill-style QA system, focused on normalizing manual test-case intake.
read→
LLM Evaluation Rubrics
The model-selection and tool-call evaluation layer behind AI-assisted QA workflows.
read→
Codex In-Browser, Safely
The safety posture for letting teammates use browser-capable AI agents.
read→