Guide · Agent skill
How to use the BrowserWing Browser Control agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The browserwing-executor skill gives an AI agent a set of commands for controlling a web browser: opening pages, clicking, typing, selecting options, reading text and taking screenshots. It works by sending HTTP requests to a BrowserWing Executor service running on your own computer at localhost:8080. This guide explains what the skill does, how to install it, how to ask for things in plain language, and what to check before letting an agent act on a real site.
BrowserWing Browser ControlMemTensor/MemOS · Apache-2.0 Download the skill Quick start
- 1Download and unzip the skill. Download the ZIP from this skill's page and unzip it so the browserwing-executor folder containing SKILL.md is in place.
- 2Place it in a skills folder. Put the folder in ~/.claude/skills/ for personal use, or in .claude/skills/ inside your project. In the Claude apps, upload the ZIP under Settings, then Capabilities, then Skills.
- 3Start the BrowserWing service. Run the BrowserWing Executor service so its API is reachable at http://localhost:8080/api/v1/executor.
- 4Set your API key. Provide your BrowserWing API key so the agent can send it as the X-BrowserWing-Key header or as a Bearer token.
- 5Check the connection. Call the help endpoint at http://localhost:8080/api/v1/executor/help to confirm the service responds and lists its commands.
- 6Ask for a task in plain language. Describe the page, the action and the output you want. Start with a low-risk task such as reading or extracting data before any clicking or form entry.
- 7Review the results. Check the extracted data or screenshot, and review any step that changed a real site before you rely on it.
What the skill lets the agent do
Once the skill is installed and the BrowserWing service is running, the agent can operate a browser for you. The skill's documentation groups the capabilities into page navigation, element interaction, data extraction, accessibility analysis, and advanced operations such as screenshots, JavaScript execution and keyboard input.
The skill also teaches the agent a workflow: list the available commands, navigate to a page, read an accessibility snapshot to find the elements on it, then click, type or extract. The agent uses the element IDs from that snapshot, such as @e1 or @e3, to point at the right controls.
- Navigate to URLs, go back, go forward and reload
- Click, type into, select from and hover over elements
- Extract text, attributes and values from elements
- Read an accessibility snapshot of the page structure
- Take a base64-encoded page screenshot and run JavaScript
- Run several operations in sequence with a batch command
Install the browserwing-executor skill
The skill is a folder whose main file is SKILL.md. Download the ZIP from the skill's page in this directory and unzip it. Then place the folder where your agent looks for skills: either your user skills folder at ~/.claude/skills/, or a project-specific .claude/skills/ folder inside the project you are working in.
If you use the Claude apps instead of Claude Code, you can upload the skill under Settings, then Capabilities, then Skills. Use the same ZIP. After installing, start a new conversation so the agent loads the skill.
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into .claude/skills/ inside your project
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Set up the BrowserWing service and API key
The skill does not drive a browser by itself. It needs a running BrowserWing Executor service that exposes its API at http://localhost:8080/api/v1/executor. You also need an API key for that service. The skill sends the key either as an X-BrowserWing-Key header or as a Bearer token in the Authorization header.
Before asking the agent for anything, check that the service answers. The help endpoint lists every command with its parameters, and the skill tells the agent to call it first when unsure.
- A BrowserWing Executor service must be running on your machine
- The API base URL is http://localhost:8080/api/v1/executor
- Authenticate with X-BrowserWing-Key: <api-key> or Authorization: Bearer <token>
curl -X GET 'http://localhost:8080/api/v1/executor/help'Trigger the skill with natural requests
You do not need to type commands yourself. Describe the task in plain language and mention the browser work you want done. The agent will read the skill and decide which API calls to make. Be specific about the site, what to look for, and what result you want back.
Good requests name the page, the action and the output format. Vague requests such as 'do something on my site' make the agent guess, which is risky when it can click and type on real pages.
- Search example.com for laptops and list the first five results with their links
- Open the signup page on my staging site and fill in the name and email fields with the test values I give you
- Take a screenshot of the pricing page and describe what it shows
- Go through the product list on this page and extract each title and link into a table
Follow a worked example
The skill's complete example covers the request 'Search for laptop on example.com and get the first 5 results'. The agent navigates to the search page, takes an accessibility snapshot to find the search box, types the query, presses Enter, waits for the results container to become visible, and extracts the matching items with their text and links.
Each of these is a separate call to the local service. If you watch the calls, you can see exactly what the agent did at each step.
- Navigate to the search page
- Read the snapshot and find the search input, for example @e3
- Type the query into that element
- Press Enter using the press-key command
- Wait for .search-results to be visible, with a 10 second timeout
- Extract .result-item elements with text and href fields
curl -X POST 'http://localhost:8080/api/v1/executor/extract' -H 'Content-Type: application/json' -d '{"selector": ".result-item", "fields": ["text", "href"], "multiple": true}'Safety notes before you let the agent act
The skill runs no scripts of its own. It uses the network, because the agent sends HTTP requests to the BrowserWing service. Those requests control a real browser, so the agent's actions have real effects on the sites it visits. The skill has no confirmation step before it clicks, types or submits.
Do not use it to log into accounts, make purchases or send messages without reviewing each step. Avoid sensitive sites such as banking or email, where the agent may see and type credentials or private page content. Use it on sites you control or have permission to automate.
- Review each action when a click or submit would change something
- Keep credentials out of prompts and out of pages the agent can read
- Prefer test accounts and staging sites when trying the skill
- Remember that page content the agent reads is sent to the agent's context
Tips for better results
Ask the agent to call the snapshot after every navigation. The skill itself says this is critical, because it provides the element IDs the agent needs. RefIDs such as @e1 are described as stable across page changes, so they are the most reliable identifier to use.
If an action depends on content that loads later, ask the agent to wait for the element first rather than acting immediately. For dialogs, the skill offers a handle-dialog command to configure how alert, confirm and prompt boxes are handled.
- Make sure the BrowserWing service is running before you start a request
- Prefer RefIDs from the snapshot; use CSS selectors or XPath when you know the structure
- Use wait with a timeout for pages that load slowly
- Use batch operations for fixed sequences of steps
- Ask for a summary format if you want extracted data in a table or list
What people use it for
- Pull a list of products, prices or listings from pages you are allowed to access, then ask the agent to summarise them
- Fill in repetitive form fields on a staging or test site during development
- Take screenshots of pages at a point in a workflow so you can check how they look
- Run a fixed sequence of navigate, click and type steps with the batch command
- Collect search results from a site into a structured summary for research
Frequently asked questions
Is the browserwing-executor skill free to use?
The source material for this skill does not state pricing or licence terms beyond the licence listed for the repository, which is Apache-2.0. You also need a running BrowserWing service and its API key, and the source does not describe how those are obtained, so check the BrowserWing documentation for that.
What do I need before installing this skill?
You need a running BrowserWing Executor service that exposes its API at http://localhost:8080/api/v1/executor, plus an API key for that service. Without the service running locally, the skill will fail. The skill itself is a folder with a SKILL.md file that you unzip into a skills directory.
Where do I put the skill folder?
For personal use, unzip it into ~/.claude/skills/. For a single project, unzip it into .claude/skills/ inside that project. In the Claude apps, you can upload the ZIP under Settings, then Capabilities, then Skills.
Does the skill run scripts or send data over the network?
The skill does not run scripts. It does use the network, because the agent sends HTTP requests to the BrowserWing service on localhost:8080, and that service controls the browser. Page content the agent reads is part of what the agent processes, so avoid sensitive sites.
Can the agent log into my accounts or make purchases?
The skill gives the agent the ability to click, type and submit on real sites, but it has no confirmation step. It is not recommended to use it for logging in, purchasing or sending messages without reviewing each action. Use test accounts or staging sites where possible.
What should I do if the agent cannot connect?
First confirm the BrowserWing service is running and reachable at localhost:8080. Then call the help endpoint with curl to see whether it responds. Check that your API key is sent in the X-BrowserWing-Key header or as a Bearer token, since the skill requires one of these.
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
This skill suits developers and analysts who already run a BrowserWing service locally and want an agent to handle repetitive browser work on sites they control or have permission to use. It is a poor fit if you do not have that service, or if you want to automate banking, email or account actions without checking each step. For those cases, pick a tool that adds confirmation steps or does not put the agent in front of your credentials.
Download the skill