Guide · Agent skill
How to use the Browser Automation agent skill (OpenFang)
Updated · written with AI from the project's own documentation and checked against our catalogue data
The Browser Automation skill gives an AI agent a playbook for controlling a web browser: opening pages, clicking buttons, typing into forms, reading page content and taking screenshots. It is written for developers, testers and people who want a summary of prices across several stores. This guide explains what the skill does, how to install it, how to trigger it, and what it needs and does not do.
Browser Automation PatternsRightNow-AI/openfang · Apache-2.0 Download the skill Quick start
- 1Download the ZIP. Download the browser-automation skill ZIP from this page.
- 2Unzip into a skills folder. Unzip it into ~/.claude/skills/ for personal use, or into .claude/skills/ in a project.
- 3Confirm your runtime tools. Make sure the OpenFang runtime provides the browser tools and memory tools the skill names, and that a browser backend is available.
- 4Make a read-only request first. Ask the agent to open a page and report what it sees, so you can confirm it works on your site.
- 5Give a bounded task. Name the site, the goal and where to stop, such as reporting prices without adding to the cart.
- 6Review before approving. When the agent reports a cart or form, check the details and only then approve the next step.
What an agent skill is and what this one makes the agent do
An agent skill is a folder of instructions that an AI agent loads to learn how to handle a particular kind of task. This skill is a prompt-only skill: it contains no executable code. It is a set of workflows, error-recovery strategies and common CSS selectors that tell the agent how to drive a browser.
The skill names six browser tools the agent should use: browser_navigate, browser_click, browser_type, browser_screenshot, browser_read_page and browser_close. It also covers a price comparison workflow that uses memory_store and memory_recall.
- Product search and purchase, stopping for your approval before checkout
- Account login, with credentials handled as described in the safety section
- Form submission, with a screenshot check before submitting
- Price comparison across several stores, followed by a summary report
Install the browser-automation skill
Download the ZIP file for the skill from this page. Then choose one of two install routes.
For a personal install, unzip the file into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. You can also upload the skill in the Claude apps under Settings → Capabilities → Skills.
- Personal: unzip into ~/.claude/skills/
- Project only: unzip into .claude/skills/ inside your project
- Claude apps: Settings → Capabilities → Skills, then upload the skill
Check the requirements before you start
The skill only describes the workflow. The browser tools must come from the host OpenFang runtime, and the skill does not include a browser automation backend. Playwright is named in the description, but the files do not contain an installation for it, so you need to provide that yourself.
The price comparison workflow also depends on the memory_store and memory_recall tools. If your runtime does not provide them, the other workflows can still apply but price comparison will not work as written.
- OpenFang runtime providing browser_navigate, browser_click, browser_type, browser_screenshot, browser_read_page and browser_close
- memory_store and memory_recall for price comparison
- A browser automation backend; Playwright is named but not installed by the skill
- Network access: the skill uses the network
Trigger the skill with natural requests
You do not need special syntax. Describe the web task in plain language and name the site or the goal. The agent can then pick the matching workflow from the skill. Be specific about what to stop for, because the skill's default is to pause before checkout.
Good requests name the outcome you want and any limit you set. For example, ask the agent to find a product on a named store, report the price, and not add anything to the cart until you confirm.
- "Open example-store.com, search for a 1 litre kettle and report the three cheapest options with prices."
- "Fill in the contact form on this page with the details below, take a screenshot, and wait for me before submitting."
- "Compare the price of this item across the three stores I list and summarise the results."
Understand the selectors and workflows the agent uses
The skill includes a CSS selector reference for basic elements, forms, navigation and e-commerce pages. Examples include #id, .class, [attr=val], input[type="email"], button[type="submit"], a[href*="checkout"] and .add-to-cart. The agent uses these to find elements, and falls back to visible text or scrolling if a selector does not match.
Each workflow is a numbered list of tool calls. The product purchase workflow reads the page, adds an item to the cart, checks the cart total, then stops and reports to you before clicking Proceed to Checkout.
- Error recovery covers missing elements, page timeouts, pop-ups, cookie banners and rate limits (the skill says to wait 30 seconds and retry)
- Wrong pages are checked with browser_read_page and the agent navigates back
Follow the safety rules
The skill's security checklist asks the agent to verify the domain before entering credentials, check for HTTPS before submitting sensitive data, report suspicious redirects, and warn about phishing signs such as misspelled domains. It says the agent must never store passwords in memory_store and must never auto-approve financial transactions.
CAPTCHAs are not solved by the skill. The agent is told to tell you instead. When a login is required, the agent asks you for credentials rather than guessing.
The skill does not run scripts, but it does use the network. Review the sites and accounts you point it at, and keep live purchases and sensitive logins under your own supervision.
- Never stores passwords in memory
- Stops and waits for your approval before checkout
- Hands CAPTCHAs back to you
- Verify domains and HTTPS before entering sensitive data
What people use it for
- Comparing prices for a single product across several online stores and getting a summary
- Walking through a form submission on a staging site to check each step
- Logging into a test account and confirming the dashboard loads
- Taking screenshots of pages for visual checks before a form is submitted
- Reading page content from a site you are allowed to automate and summarising it
Frequently asked questions
Is the browser automation skill free to use?
The source material does not state a price or licence terms beyond the Apache-2.0 licence listed for the item. Check the licence file in the repository for the terms that apply.
What do I need to install before using this skill?
You need the OpenFang runtime with the six browser tools and the memory tools named in the skill, plus a browser automation backend. Playwright is named in the description, but the files do not include its installation.
Where do I put the browser automation skill folder?
Unzip the downloaded ZIP into ~/.claude/skills/ for personal use, or into .claude/skills/ inside a project. You can also upload it in the Claude apps under Settings → Capabilities → Skills.
Does the skill include working browser code?
No. The skill is a set of instructions and CSS selectors. The browser tools must be provided by the host OpenFang runtime.
Can the agent buy things for me automatically?
The skill tells the agent to stop and report before clicking checkout and to wait for your approval. It also says never to auto-approve financial transactions. Even so, the skill is prompt guidance only, so the agent may still act incorrectly on live sites, and you should supervise purchases.
What should I do if the agent hits a CAPTCHA or a login wall?
The skill says the agent cannot solve CAPTCHAs and should inform you. For logins, it asks you for credentials rather than storing them, so you will need to supply them when prompted.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you are a developer, tester or price-comparer who wants an agent to drive a browser through repetitive, supervised web tasks, and you already have OpenFang and a browser backend. Pick something else if you want unattended purchases, logins to sensitive accounts, or scraping of sites whose terms forbid automated access, since the skill is only guidance and does not contain browser code.
Download the skill