phuetz/code-buddy/web-automate

Browser automation with Playwright (Python)

Drive a real browser programmatically with the open-source playwright Python library: navigate, screenshot, scrape JS-rendered content, fill and submit forms. Work in visible steps and verify each result before reporting.

For one-off, point-and-click control prefer the built-in browser/computer-use

tools. Use this skill when the user wants a scripted, reproducible browser

task (scrape N pages, screenshot a flow, fill a form) written as Python.

Preflight (extras-tier — lib + browser binary)

python3 - <<'PY'
try:
    import playwright  # noqa: F401
except ModuleNotFoundError:
    raise SystemExit(
        "playwright not installed. Run `npm run prepare:python:extras` in cowork/ "
        "(installs playwright + fetches the Firefox binary)."
    )
print("playwright OK")
PY

If the browser binary is missing at runtime, fetch it once: python3 -m playwright install firefox (honors PLAYWRIGHT_BROWSERS_PATH, which the app points at the bundled resources/python/<plat>/ms-playwright).

Workflow (one step at a time)

1. Plan the target URL and the action (screenshot / scrape / form).
2. Write a short Python script (heredoc via bash) using the sync API,

launching Firefox headless with --no-sandbox not required for Firefox.

1. Run it; capture the artifact (PNG / extracted text).
2. Verify: assert the screenshot file is non-empty, or the scraped text is

non-empty; print a short confirmation.

1. Report the artifact path / extracted data and a one-line summary.

Conventions

• Always run headless (headless=True) — the app has no display.
• Set a realistic user_agent and a viewport; wait_until="networkidle" for JS pages.
• Use page.wait_for_selector(...) instead of fixed sleeps.
• Always browser.close() in a finally / context manager.
• Respect the target site's terms; this is for automating sites the user is

authorized to use, not for bypassing access controls.

Example — screenshot + scrape a page

python3 - <<'PY'
import os
from playwright.sync_api import sync_playwright
 
URL = "https://example.com"
out = "page.png"
with sync_playwright() as p:
    browser = p.firefox.launch(headless=True)
    page = browser.new_page(viewport={"width": 1280, "height": 800})
    page.goto(URL, wait_until="networkidle")
    title = page.title()
    heading = page.inner_text("h1")
    page.screenshot(path=out, full_page=True)
    browser.close()
 
assert os.path.getsize(out) > 0, "screenshot is empty"
print(f"title={title!r} h1={heading!r}; wrote {out} ({os.path.getsize(out)} bytes)")
PY

Optional: Camoufox stealth backend

For sites that fingerprint/block automation, camoufox (a hardened Firefox) is a drop-in replacement; install with prepare:python:extras then python3 -m camoufox fetch:

from camoufox.sync_api import Camoufox
with Camoufox(headless=True) as browser:
    page = browser.new_page()
    page.goto("https://example.com")
    print(page.title())

Camoufox is optional — plain Playwright Firefox is the default.