QA Pipeline

An AI-powered QA platform built on OpenCode’s agent infrastructure that automates the entire testing lifecycle from requirements to reports.

🎯 Overview

QA Pipeline takes a web application URL (and optionally a PRD), walks the app like a real user using Playwright MCP, generates executable Playwright test scripts, and produces a severity-ranked QA report with a ship/no-ship verdict.

Key Features

• Automated Requirement Analysis - Parse PRDs into structured requirements
• Exploratory Testing - Discover pages, forms, and user flows
• Playwright MCP Integration - Capture DOM, screenshots, console logs, network requests
• Test Generation - Generate executable Playwright tests from requirements + DOM
• Test Execution - Run tests and capture evidence (traces, videos, screenshots)
• Comprehensive Reporting - Generate QA reports with verdict and recommendations

Quick Start

Prerequisites

• Node.js 20+
• Playwright (auto-installed via npm install && npx playwright install chromium)
• OpenCode CLI (optional — only needed for Skills 2 & 4 with QA_AGENT=opencode)

Quick Demo (no agent required)

git clone https://github.com/SamuelMtzA/qa-pipeline.git
cd qa-pipeline
npm install && npx playwright install chromium

# Run the pipeline against a live demo app (QA_AGENT=none by default)
node orchestrator.js https://demo.playwright.dev/todomvc --prd examples/prd-todo.md --skip-execution

Expected output:

[1/6] Requirement Analysis
✓ Parsed 4 features from examples/prd-todo.md

[3/6] Playwright Capture
✓ Capture complete: 1 page, 3 selectors, 0 console errors

[2/6] Exploratory Testing
✓ Exploration complete: 1 page, 4 elements, 1 flow discovered

[4/6] Test Generation
✓ Test generation complete: 9 tests (2 verified, 7 unverified), 22% coverage

[5/6] Test Execution
⚠ Skipped (--skip-execution)

[6/6] Reporting
✓ Report generated: .qa-workspace//09-report/qa-report.md
   Verdict: INCONCLUSIVE | Health: 7.5/10

All 6 skills run without any agent runtime installed. Skill 3 (capture) runs before Skill 2 (exploration) in heuristic mode, since exploration analyzes capture data. Skills 2 and 4 use heuristic/template-based engines instead of LLM agents. For enhanced analysis (multi-viewport, complex flow detection, LLM-generated tests):

QA_AGENT=opencode node orchestrator.js https://demo.playwright.dev/todomvc --prd examples/prd-todo.md

Artifacts are written to .qa-workspace//:

• 00-config.json — run config with blast_radius
• requirements/requirements.json — parsed PRD features
• playwright-output/ — DOM snapshot, screenshots, console logs, network logs, selectors
• 09-report/qa-report.md + qa-report.json — verdict, health score, coverage, recommendations

Basic Usage

# Run the pipeline
node orchestrator.js https://example.com

# With PRD
node orchestrator.js https://example.com --prd docs/PRD.md

# Using npm script