web-browse

name	web-browse
description	Browse the web — take screenshots, extract text, inspect page elements, and execute JavaScript using Playwright. Use when visiting a webpage, taking a screenshot, extracting text, discovering page structure, or running JS on a site.

Web Browse Skill

Load URLs in a headless browser and extract visual, textual, or structural content.

Actions

Screenshot — Visual capture

python3 {baseDir}/browse.py screenshot <url> [options]

Options:

• --output <path> — Save screenshot to file (default: /tmp/screenshot-<timestamp>.png)
• --width <px> — Viewport width (default: 1280)
• --height <px> — Viewport height (default: 800)
• --full-page — Capture full scrollable page, not just viewport
• --wait <ms> — Wait time after page load before capture (default: 2000)
• --selector <css> — Screenshot only a specific element

Text — Content extraction

python3 {baseDir}/browse.py text <url> [options]

Options:

• --output <path> — Save text to file (default: stdout)
• --selector <css> — Extract text from specific element only
• --wait <ms> — Wait time after page load (default: 2000)
• --links — Include href URLs in output
• --max-length <chars> — Truncate output (default: 50000)

Inspect — Element discovery

Discover all interactive elements on a page — buttons, links, inputs, forms. Useful for debugging Playwright automation or understanding a page's structure.

python3 {baseDir}/browse.py inspect <url> [options]

Options:

• --wait <ms> — Wait time after page load (default: 2000)
• --width <px> — Viewport width (default: 1280)
• --console — Capture and display browser console output
• --screenshot <path> — Also save a full-page screenshot
• --json — Output element data as JSON (for programmatic use)

Execute — Run JavaScript

Execute a JavaScript expression on a loaded page and return the result.

python3 {baseDir}/browse.py execute <url> "<script>" [options]

Options:

• --wait <ms> — Wait time after page load (default: 2000)

Examples

# Screenshot a web page
python3 {baseDir}/browse.py screenshot "https://example.com" --output /tmp/example.png

# Full-page screenshot
python3 {baseDir}/browse.py screenshot "https://example.com" --full-page --output /tmp/site.png

# Extract text from a page
python3 {baseDir}/browse.py text "https://example.com/article" --max-length 10000

# Extract text with links
python3 {baseDir}/browse.py text "https://github.com/example/repo" --links

# Discover all interactive elements on a page
python3 {baseDir}/browse.py inspect "https://example.com/form"

# Inspect with console capture and screenshot
python3 {baseDir}/browse.py inspect "https://example.com" --console --screenshot /tmp/inspect.png

# Inspect and get JSON output for scripting
python3 {baseDir}/browse.py inspect "https://example.com/form" --json

# Execute JavaScript to extract specific data
python3 {baseDir}/browse.py execute "https://example.com" "document.querySelectorAll('img').length"

# Get page metadata via JS
python3 {baseDir}/browse.py execute "https://example.com" \
  "JSON.stringify({title: document.title, links: document.querySelectorAll('a').length})"

Reconnaissance Pattern

For debugging automation or understanding unfamiliar pages:

1. Inspect — discover all elements:

   python3 {baseDir}/browse.py inspect "https://platform.com/upload" --console --screenshot /tmp/recon.png
   

2. Read the screenshot — verify visual state matches expectations
3. Identify selectors — from the inspect output, find the right buttons/inputs
4. Execute — test selectors with JavaScript:

   python3 {baseDir}/browse.py execute "https://platform.com/upload" \
     "document.querySelector('button[type=submit]')?.innerText"
   

Notes

• Uses headless Chromium via Playwright (no GUI required)
• Screenshots are PNG format, suitable for vision analysis via the Read tool
• Text extraction strips HTML tags and returns clean content
• JavaScript renders fully before capture (configurable wait time)
• Navigation falls back from networkidle to load on timeout
• No authentication — for authenticated sites, use platform-specific skills

Troubleshooting

Page load timeout

Some pages are slow or have infinite network activity. Try increasing --wait or the page may be blocking headless browsers.

Selector not found

Verify the selector with inspect first. Dynamic pages may load elements after the initial networkidle state — increase --wait.

Console shows errors

Use inspect --console to see what the page is complaining about. Common: CORS errors (expected for cross-origin resources), missing APIs (page expects auth).