browser-automation

Overview of browser-automation skill

What browser-automation is built to do

The browser-automation skill helps an AI agent design and implement Playwright-based workflows for real browser tasks: scraping structured data, filling forms, downloading files, capturing screenshots or PDFs, handling sessions, and working with JavaScript-heavy pages. It is not a generic “open a browser” prompt; it gives the agent automation patterns, Playwright API guidance, extraction recipes, and helper scripts that reduce guesswork.

Best-fit users and projects

Install this skill if you want Claude to help build browser automation scripts for data collection, operational workflows, or repeatable web tasks. It is especially useful for engineers, data teams, growth teams, and operations users who already know the target website and need a reliable automation plan or script skeleton. If your goal is end-to-end testing, use a testing-focused skill such as playwright-pro instead.

What makes this browser-automation skill useful

The main value is its practical bias toward production automation rather than demos. The repository includes references for playwright_browser_api.md, data_extraction_recipes.md, and anti_detection_patterns.md, plus scripts for generating scraping and form-fill skeletons. This gives the agent a better starting point for selectors, pagination, browser contexts, screenshots, file uploads, error handling, and session reuse.

Important constraints before adoption

The skill assumes Playwright-style browser automation and is strongest with Python async examples. It will not bypass legal, access-control, paywall, or terms-of-service restrictions for you. Anti-detection material should be used to reduce false bot flags for legitimate automation, not to abuse protected services. For sites with aggressive bot protection, fragile markup, or frequent UI changes, expect iteration.

How to Use browser-automation skill

browser-automation install and first files to read

Install with:

npx skills add alirezarezvani/claude-skills --skill browser-automation

Then inspect the skill source at:

engineering/skills/browser-automation

Read SKILL.md first to understand trigger conditions and scope. Next, read references/playwright_browser_api.md for browser contexts, launch options, downloads, screenshots, and network handling. Use references/data_extraction_recipes.md when the task involves listings, tables, article pages, job boards, or product cards. Review references/anti_detection_patterns.md only when reliability is affected by bot-detection signals.

Inputs the skill needs for better output

A weak browser-automation usage prompt says: “Scrape this website.” A stronger prompt gives the target URL, exact fields, page types, login/session needs, pagination pattern, output format, rate limits, and failure behavior.

Example prompt:

“Use the browser-automation skill to create a Python Playwright scraper for https://example.com/jobs. Extract job title, company, location, salary, detail URL, and posting date. Results are paginated with a Next button. Save JSON Lines. Use browser contexts, avoid fixed sleeps, include retries for missing selectors, and add comments where I must update CSS selectors.”

This gives the agent enough constraints to choose extraction patterns, output structure, waiting strategy, and error handling instead of producing a generic script.

Practical browser-automation usage workflow

Start by asking the agent for a plan, not code. Have it identify page states: landing page, login, search results, detail page, pagination, download step, or confirmation page. Then provide sample HTML snippets, screenshots, or inspected selectors when possible. After the plan, ask for a minimal Playwright script and a selector map separated from the workflow logic.

For scraping tasks, consider using scripts/scraping_toolkit.py as a skeleton generator. For forms, use scripts/form_automation_builder.py with a JSON field specification. For reliability review, run or adapt scripts/anti_detection_checker.py against generated Playwright code to catch obvious issues such as default viewport, missing user agent, no throttling, or weak session handling.

Tips that improve script quality

Ask for browser contexts instead of repeatedly launching browsers. Prefer Playwright auto-waiting and locator logic over sleep() calls. Request structured logging, screenshots on failure, and clear output schemas. For multi-step form automation, include field types such as text, email, select, checkbox, radio, file, and click, plus the expected success condition after submission.

browser-automation skill FAQ

Is browser-automation for Browser Automation or testing?

It is for Browser Automation workflows, not test suites. Use this skill when the browser is a tool for completing work: scraping, form submission, screenshots, downloads, session-based workflows, and structured extraction. If you need assertions, fixtures, CI test reporting, or E2E test design, a Playwright testing skill is a better fit.

Can beginners use this browser-automation skill?

Yes, but beginners should provide more context and ask for explanations. The skill can generate scripts and plans, but you still need to understand URLs, selectors, credentials handling, and local Playwright setup. If you cannot inspect a page or describe the fields you need, output quality will be limited.

How is it better than an ordinary prompt?

An ordinary prompt may produce a simple Playwright script, but it often misses operational details: session isolation, pagination, structured extraction, retries, screenshots on failure, selector maps, downloads, and anti-detection risk checks. The browser-automation skill gives the agent a domain-specific playbook and supporting files to produce more usable automation.

When should I not install this skill?

Do not install it if your work is only API automation, load testing, security testing, or browser-based test automation. It is also a poor fit when the target site explicitly forbids automation and you cannot obtain permission. If the workflow can be solved with a public API, prefer the API because it is usually more stable than browser scraping.

How to Improve browser-automation skill

Give the browser-automation skill stronger task specs

The most common failure mode is underspecified intent. Improve prompts by including: target pages, required fields, selectors if known, authentication flow, pagination behavior, output format, maximum pages or records, retry policy, and what counts as success. For form workflows, provide a JSON-like list of fields with selector, type, and value.

Iterate from generated script to reliable workflow

After the first output, test on one page or one form submission before scaling. Ask the agent to revise based on actual errors: timeout messages, missing selector names, screenshots, console logs, network failures, or changed redirects. A useful follow-up is: “Here is the failing selector and the page snippet; update the locator strategy and add a fallback.”

Reduce fragility with better selectors and state handling

Prefer stable attributes such as data-testid, accessible roles, labels, and semantic containers over brittle generated CSS paths. Ask the agent to separate selector configuration from workflow code so future site changes are easier to fix. For logged-in workflows, request browser storage state or cookie handling rather than redoing login on every run.

Improve reliability without overusing anti-detection

Use anti-detection guidance only where appropriate: realistic viewport, locale, user agent, pacing, and session reuse can improve legitimate automation reliability. Do not treat these patterns as a guarantee against advanced bot defenses. For high-value workflows, add monitoring, respectful rate limits, failure screenshots, and a manual review path before running at scale.