UI Test — Plain English E2E Testing (🧪)

Describe your UI tests in plain English. The agent figures out how to find buttons, navigate the app flow, and execute each step — no selectors or code needed. It screenshots every major step, stitches them into a walkthrough video, and DMs you the result with a pass/fail report.

Made in 🤠 Texas ❤️ PlebLab

Workflow

1. Create — User names a test and describes steps in plain English
2. Run — Agent opens the clawd browser, executes each step, screenshots each action
3. Export — Agent generates a Playwright .spec.ts from the verified steps
4. CI/CD — User drops the script into their test suite, runs with npx playwright test

Agent Execution Flow

When running a test:

1. Load test definition: node scripts/ui-test.js get "<name>"
2. Start clawd browser: browser action=start profile=clawd
3. Navigate to the test URL
4. For each plain English step: a. Interpret what the user means (click, type, assert, wait, etc.) b. Use browser action=snapshot to see the page c. Use browser action=act with the appropriate request (click/type/press/etc.) d. Take a screenshot after each step e. Record what selector/ref was used and whether it passed
5. Save run results: node scripts/ui-test.js save-run "<name>" passed=true/false
6. Report results to user with pass/fail per step

When exporting to Playwright:

1. Load the test definition and most recent successful run
2. Map each plain English step to Playwright API calls based on what worked during execution
3. Generate a .spec.ts file with proper imports, test structure, and assertions
4. Save to the user's project or a specified output path

Step Interpretation Guide

The agent should interpret plain English steps like:

User says	Browser action	Playwright equivalent
"click the Sign In button"	act: click ref="Sign In button"	page.getByRole('button', {name: 'Sign In'}).click()
"type hello@test.com in the email field"	act: type ref="email" text="hello@test.com"	page.getByLabel('Email').fill('hello@test.com')
"verify the dashboard shows Welcome"	snapshot + check text	expect(page.getByText('Welcome')).toBeVisible()
"wait for the page to load"	act: wait	page.waitForLoadState('networkidle')
"click the hamburger menu"	act: click (find menu icon)	page.getByRole('button', {name: 'menu'}).click()
"scroll down"	act: evaluate fn="window.scrollBy(0,500)"	page.evaluate(() => window.scrollBy(0, 500))
"check the Remember Me checkbox"	act: click ref="Remember Me"	page.getByLabel('Remember Me').check()
"select 'USD' from the currency dropdown"	act: select values=["USD"]	page.getByLabel('Currency').selectOption('USD')
"take a screenshot"	browser action=screenshot	page.screenshot({path: 'step-N.png'})
"verify URL contains /dashboard"	check current URL	expect(page).toHaveURL(/dashboard/)

Commands

Run via: node ~/workspace/skills/ui-test/scripts/ui-test.js <command>

Command	Description
create <name> [url]	Create a new test
add-step <name> <step>	Add a plain English step
set-steps <name> <json>	Replace all steps
set-url <name> <url>	Set the test URL
get <name>	Show test definition
list	List all tests
remove <name>	Delete a test
save-run <name> ...	Save execution results
runs [name]	Show run history
export <name> [outfile]	Export as Playwright script

Export Format

Generated Playwright files include:

• Proper TypeScript imports
• test.describe block with test name
• test.beforeEach with navigation to base URL
• Each step as a sequential action with comments showing the original English
• Assertions where the user said "verify", "check", "should", "expect"
• Screenshots on failure

Screenshots & Video

During test execution, the agent should:

1. Before each step: take a screenshot → save as step-NN-before.jpg
2. After each step: take a screenshot → save as step-NN-after.jpg
3. On failure: take a screenshot → save as step-NN-FAIL.jpg

Screenshots are saved to: ~/.ui-tests/runs/<slug>-<timestamp>/

After the run completes, generate a walkthrough video:

ffmpeg -framerate 1 -pattern_type glob -i '~/.ui-tests/runs/<folder>/step-*.jpg' \
  -vf "scale=1280:-2" -c:v libx264 -pix_fmt yuv420p -y output.mp4

Then send the video to the chat.

Storage

• Test definitions: ~/.ui-tests/<slug>.json
• Run history: ~/.ui-tests/runs/<slug>-<timestamp>/run.json
• Screenshots: ~/.ui-tests/runs/<slug>-<timestamp>/step-*.jpg
• Video: ~/.ui-tests/runs/<slug>-<timestamp>/walkthrough.mp4
• Exported scripts: user-specified path or ./tests/<slug>.spec.ts