qa - SKILL.md Agent Skill

name: qa license: MIT description: >- Browser-based QA verification. Launches a real browser, navigates the app, clicks buttons, fills forms, and tests user flows. Works as a standalone skill or as a phase end condition in campaigns. Requires Playwright (optional dependency, graceful skip if not installed). user-invocable: true auto-trigger: false trigger_keywords: - qa - test the app - click through - does it work - browser test effort: high

/qa — Browser QA Verification

Dependency: Playwright

/qa requires Playwright. It's an optional dependency.

If Playwright is installed: full browser QA works. If Playwright is NOT installed: the skill offers to install it, or falls back to /live-preview (screenshot-only verification).

Detection:

npx playwright --version 2>/dev/null

Installation (if user agrees):

npm install -D playwright
npx playwright install chromium

Only installs Chromium (smallest download, ~150MB). Not Firefox or WebKit unless the user asks for cross-browser testing.

/do setup integration: During setup, if the project is a web app (has React, Next.js, Vue, Svelte, or HTML files), offer to install Playwright: "I see this is a web project. Want to enable browser QA testing? This installs Playwright (~150MB) for interaction testing. (y/n)"

If they say no, /qa falls back to /live-preview. No pressure.

When to Use

After building a feature (verify it actually works in a browser)
As a phase end condition: "QA verification passes for [flow]"
When /do routes "qa", "test the app", "does it work", "click through it"
When /create-app campaigns reach the verification phase
After /live-preview shows something renders but you need to verify interactions

Protocol

Step 1: DISCOVER

Before testing, understand what to test:

Read the project's routes/pages (from file tree, router config, or package.json scripts)
Read the PRD or campaign file (if exists) for expected user flows
Identify testable flows:
- Page loads and renders (baseline)
- Navigation between pages
- Form submissions
- Button click handlers
- Auth flows (login, logout, protected routes)
- CRUD operations (create, read, update, delete)
- Error states (invalid input, network errors)

If no PRD or campaign exists, ask: "What should I test? Give me 1-3 user flows."

Step 2: START THE APP

Before testing, the app needs to be running:

Check if a dev server is already running (try curl localhost:3000, 5173, 8080)
If not running, check package.json for start/dev scripts
Start it: npm run dev or equivalent, in background
Wait for the server to be ready (poll the health endpoint or main URL)
If the app won't start, report the error and stop. Don't test a broken app.

Track whether the agent started the server. If so, kill it on completion.

Step 3: TEST

For each flow identified in Step 1, write and run a Playwright script:

const { chromium } = require('playwright');

(async () => {
  const browser = await chromium.launch({ headless: true });
  const page = await browser.newPage();

  // Navigate
  await page.goto('http://localhost:3000');

  // Verify page loaded
  const title = await page.title();

  // Test interactions
  await page.click('button[data-testid="add-todo"]');
  await page.fill('input[name="title"]', 'Test todo');
  await page.click('button[type="submit"]');

  // Verify result
  const todoText = await page.textContent('.todo-item:last-child');

  // Screenshot for evidence
  await page.screenshot({ path: '.planning/screenshots/qa-flow-1.png' });

  await browser.close();
})();

For each test:

Navigate to the relevant page
Perform the user action (click, fill, submit)
Verify the expected outcome (element appears, text changes, navigation occurs)
Take a screenshot as evidence
Log: PASS or FAIL with description

Step 4: REPORT

Write results to .planning/qa-report-{date}.md:

# QA Report: {App Name or Feature}

> Date: {ISO date}
> Flows tested: {N}
> Passed: {N}
> Failed: {N}
> Screenshots: .planning/screenshots/qa-*.png

## Results

### Flow 1: {description}
- Steps: {what was done}
- Expected: {what should happen}
- Actual: {what did happen}
- Result: PASS / FAIL
- Screenshot: {path}
- Notes: {any observations}

### Flow 2: ...

Step 5: CAMPAIGN INTEGRATION

When running as a phase end condition:

The campaign file can specify QA conditions:

| 3 | qa_verify | /qa passes for: add todo, complete todo, delete todo |

/qa reads the condition, runs those specific flows, and reports pass/fail. The phase is complete only if all specified flows pass.

Cookie and Auth Support

For apps with authentication:

First run the auth flow: navigate to login, fill credentials, submit
Save the browser context (cookies + localStorage state)
Use the saved context for all subsequent tests
This means authenticated flows work without re-logging-in per test

Test credentials should come from .env.example or the campaign file. NEVER read from .env (protected by the hook). Use test accounts only.

Fallback: No Playwright

If Playwright isn't installed and the user declines installation:

Fall back to /live-preview (screenshot-only)
Report: "Browser QA unavailable (Playwright not installed). Visual verification only."
Take screenshots of each page that would have been tested
Mark interaction tests as SKIPPED, visual tests as PASS/FAIL

What /qa Does NOT Do

Install Playwright without asking
Test in production (localhost only, unless user explicitly provides a URL)
Replace unit/integration tests (this is user-flow testing, not code testing)
Run on every edit (too expensive — invoked explicitly or as phase end condition)
Access .env files (uses .env.example or test credentials from campaign)

Quality Gates

Every tested flow has all fields filled (steps, expected, actual, result)
Screenshots are taken for every flow (pass or fail)
Failed flows have enough detail to reproduce the issue
The app is actually running before tests execute (not testing a dead server)

Fringe Cases

Playwright not installed and user declines: Fall back to /live-preview. Mark all interaction tests as SKIPPED in the report. Visual-only verification still runs.

Dev server won't start: Report the startup error and stop. Do not attempt to test a server that isn't running. Suggest the user fix the startup error first.

No routes or pages discoverable: Ask the user for 1-3 flows to test. Do not guess at routes.

No UI (API-only project): Report "No UI detected — /qa requires a browser-accessible interface. Use typecheck and unit tests for API verification." Then stop gracefully.

If .planning/screenshots/ does not exist: Create it before saving screenshots. If .planning/ doesn't exist, save screenshots to a qa-screenshots/ directory in the project root and note the path in the report.

Codex App Artifact Manifest

After saving screenshots, videos, rendered PDFs, or QA reports in Codex, register each durable artifact so the Codex app/browser workflow can find it later:

node scripts/codex-app-artifacts.js record --workflow qa --kind screenshot --path ".planning/screenshots/qa-flow-1.png" --status pass

The manifest lives at .planning/artifacts/codex-app-evidence.jsonl.

Before reporting QA as complete in Codex, verify the manifest points at real files:

node scripts/codex-app-artifacts.js verify --require-artifacts

Contextual Gates

Disclosure: May start a dev server; saves screenshots and report to .planning/. States server ownership before starting. Reversibility: amber — creates report and screenshots; dev server stopped only if this skill started it. Delete generated files to undo. Trust gates: Any. Familiar (5+) to run against external/production URLs.

Exit Protocol

---HANDOFF---
- QA Report: .planning/qa-report-{date}.md
- Flows tested: {N}
- Passed: {N} | Failed: {N} | Skipped: {N}
- Screenshots: .planning/screenshots/qa-*.png
- Server: {started by agent (killed) | was already running (left running)}
- Reversibility: amber — delete `.planning/qa-report-{date}.md` and `screenshots/qa-*.png` to undo
---