Playwright Skill

Playwright Skill - Browser Automation

Automate browser tasks: navigate, click, type, extract data, take screenshots, and more.

First-Time Setup

pip install playwright
playwright install chromium

Core Concepts

Sessions

Sessions persist browser state (cookies, localStorage) across commands. Use named sessions for different accounts/sites:

# Default session
python3 ~/.claude/skills/playwright-skill/playwright_skill.py open https://example.com

# Named session for a specific site
python3 ~/.claude/skills/playwright-skill/playwright_skill.py open https://github.com --session github

Sessions are saved to ~/.claude/skills/playwright-skill/sessions/

Headless vs Visible

By default, browser runs headless (invisible). Use --visible to see what's happening:

python3 ~/.claude/skills/playwright-skill/playwright_skill.py open https://example.com --visible

Commands

Navigation

# Open a URL
python3 ~/.claude/skills/playwright-skill/playwright_skill.py open URL [--session NAME] [--visible]

# Wait for element to appear
python3 ~/.claude/skills/playwright-skill/playwright_skill.py wait SELECTOR [--timeout MS] [--session NAME]

# Scroll page
python3 ~/.claude/skills/playwright-skill/playwright_skill.py scroll [--direction up|down] [--amount PIXELS] [--session NAME]

Interaction

# Click an element
python3 ~/.claude/skills/playwright-skill/playwright_skill.py click SELECTOR [--session NAME]

# Type text into input
python3 ~/.claude/skills/playwright-skill/playwright_skill.py type SELECTOR "text to type" [--clear] [--session NAME]

# Execute JavaScript
python3 ~/.claude/skills/playwright-skill/playwright_skill.py eval "document.title" [--session NAME]

Data Extraction

# Extract text from element
python3 ~/.claude/skills/playwright-skill/playwright_skill.py extract SELECTOR [--session NAME]

# Extract attribute
python3 ~/.claude/skills/playwright-skill/playwright_skill.py extract "a.link" --attr href [--session NAME]

# Extract from all matching elements
python3 ~/.claude/skills/playwright-skill/playwright_skill.py extract "li.item" --all [--session NAME]

# Get page HTML
python3 ~/.claude/skills/playwright-skill/playwright_skill.py html [--selector SELECTOR] [--output FILE] [--session NAME]

Screenshots & PDF

# Screenshot current page
python3 ~/.claude/skills/playwright-skill/playwright_skill.py screenshot [--output FILE] [--session NAME]

# Screenshot with URL
python3 ~/.claude/skills/playwright-skill/playwright_skill.py screenshot https://example.com --output example.png

# Full-page screenshot
python3 ~/.claude/skills/playwright-skill/playwright_skill.py screenshot --full-page --output full.png

# Save as PDF
python3 ~/.claude/skills/playwright-skill/playwright_skill.py pdf https://example.com --output page.pdf

Cookies & Sessions

# List all cookies
python3 ~/.claude/skills/playwright-skill/playwright_skill.py cookies [--session NAME]

# Get specific cookie
python3 ~/.claude/skills/playwright-skill/playwright_skill.py cookies --name session_id [--session NAME]

# Set cookie
python3 ~/.claude/skills/playwright-skill/playwright_skill.py cookies --set "name=value" --domain example.com [--session NAME]

# List all sessions
python3 ~/.claude/skills/playwright-skill/playwright_skill.py sessions

# Close session (saves state)
python3 ~/.claude/skills/playwright-skill/playwright_skill.py close [--session NAME]

# Close all sessions
python3 ~/.claude/skills/playwright-skill/playwright_skill.py close --all

Selectors

Playwright supports multiple selector types:

# CSS selectors
python3 ~/.claude/skills/playwright-skill/playwright_skill.py click "button.submit"
python3 ~/.claude/skills/playwright-skill/playwright_skill.py click "#login-form input[type=email]"

# Text selectors
python3 ~/.claude/skills/playwright-skill/playwright_skill.py click "text=Sign In"
python3 ~/.claude/skills/playwright-skill/playwright_skill.py click "text=Submit"

# XPath
python3 ~/.claude/skills/playwright-skill/playwright_skill.py click "xpath=//button[@type='submit']"

# Combining
python3 ~/.claude/skills/playwright-skill/playwright_skill.py click "form >> text=Submit"

Examples

Login Flow (Manual First Time)

# Open login page visibly
python3 ~/.claude/skills/playwright-skill/playwright_skill.py open https://example.com/login --visible --session mysite

# ... manually log in while browser is visible ...

# Close to save session
python3 ~/.claude/skills/playwright-skill/playwright_skill.py close --session mysite

# Future runs use saved session (logged in)
python3 ~/.claude/skills/playwright-skill/playwright_skill.py open https://example.com/dashboard --session mysite

Scrape a List

# Open page
python3 ~/.claude/skills/playwright-skill/playwright_skill.py open https://news.ycombinator.com

# Extract all headlines
python3 ~/.claude/skills/playwright-skill/playwright_skill.py extract ".titleline a" --all

# Extract with links
python3 ~/.claude/skills/playwright-skill/playwright_skill.py extract ".titleline a" --attr href --all

Fill a Form

python3 ~/.claude/skills/playwright-skill/playwright_skill.py open https://example.com/form --session formtest
python3 ~/.claude/skills/playwright-skill/playwright_skill.py type "input[name=email]" "user@example.com" --session formtest
python3 ~/.claude/skills/playwright-skill/playwright_skill.py type "input[name=message]" "Hello world" --session formtest
python3 ~/.claude/skills/playwright-skill/playwright_skill.py click "button[type=submit]" --session formtest

Screenshot a Page

python3 ~/.claude/skills/playwright-skill/playwright_skill.py screenshot https://example.com --output example.png --full-page

Run JavaScript

# Get page title
python3 ~/.claude/skills/playwright-skill/playwright_skill.py eval "document.title"

# Get scroll position
python3 ~/.claude/skills/playwright-skill/playwright_skill.py eval "window.scrollY"

# Count elements
python3 ~/.claude/skills/playwright-skill/playwright_skill.py eval "document.querySelectorAll('a').length"

Output

All commands output JSON for easy parsing.

Requirements

• Python 3.9+
• pip install playwright
• playwright install chromium

Tips

• First login: Use --visible to manually log in, then save session for headless reuse
• Flaky selectors: Use wait command before interacting with dynamic content
• Rate limiting: Add delays between actions with eval "await new Promise(r => setTimeout(r, 2000))"
• Debugging: Take screenshots to see what went wrong
• Multiple accounts: Use different session names for each account

Security Notes

• Session files contain authentication cookies - keep sessions/ directory secure
• Don't share session files - they can grant account access
• Clear sessions when done with sensitive sites: close --session NAME