Browserbase Skills: Claude Code Web Automation Plugin System

Browserbase Skills is a plugin and skill system for Claude Code that delivers comprehensive web automation capabilities through 13 specialized skills. Whether you need to browse websites, fill forms, solve CAPTCHAs, run adversarial UI tests, or build self-improving browser automation, Browserbase Skills Claude Code web automation provides a unified CLI-driven interface that works identically in both local and remote browser environments. With over 2,185 GitHub stars and 320 stars per day during its trending peak, this project has quickly become the go-to solution for giving AI coding agents real browser control.

Understanding the Architecture

The architecture diagram above illustrates how Browserbase Skills integrates with Claude Code and the Browserbase cloud platform. Let us break down each component and how they interact:

Claude Code Agent At the top of the stack sits Claude Code, the AI coding assistant that serves as the orchestrator. When a user asks Claude to browse a website, test a UI, or extract data, Claude dispatches commands through the skill system rather than attempting raw HTTP requests. This means the agent gets full browser rendering, JavaScript execution, and interactive capabilities that simple fetch operations cannot provide.

Skill Registry and Plugin System The skill registry acts as the central dispatcher. When installed via npx skills add browserbase/skills or the /plugin command in Claude Code, all 13 skills become available as natural language invocations. Each skill has its own SKILL.md file that defines triggers, allowed tools, and compatibility requirements. The registry ensures that Claude knows which skill to activate based on the user’s intent – whether that is browsing, testing, debugging, or researching.

Daemon-Based Browse CLI The browse CLI is the workhorse of the system. It runs as a persistent daemon process that manages browser sessions, handles page navigation, and provides structured output through accessibility tree snapshots. The daemon auto-starts on first command and supports both local Chrome instances and remote Browserbase cloud sessions. Key commands include browse open, browse snapshot, browse click, browse fill, and browse stop.

Browserbase Cloud API For sites with bot detection, CAPTCHAs, or geo-restrictions, the system seamlessly switches to Browserbase’s cloud infrastructure. This provides anti-bot stealth mode with custom Chromium builds, automatic CAPTCHA solving for reCAPTCHA and hCaptcha, residential proxies across 201 countries, and session persistence through cookie contexts. The transition between local and remote is a single command: browse env remote.

Key Insight: Browserbase Skills provides residential proxies in 201 countries with automatic CAPTCHA solving, making it possible for AI agents to access protected websites that would normally block automated browsing.

The 13 Skills: Core and Advanced

The ecosystem diagram organizes all 13 skills into two tiers. Here is a detailed look at each:

Core Skills (4)

Skill	Purpose	Key Capability
browser	Web browser automation	Full CLI-driven browsing with accessibility tree snapshots, element interaction, and session management
fetch	Static page retrieval	Fetch HTML or JSON without a browser session – inspect status codes, headers, and follow redirects
search	Web search	Return structured search results (titles, URLs, metadata) without launching a browser
browserbase-cli	Platform management	Use the official bb CLI for Browserbase Functions, sessions, projects, contexts, extensions, and dashboard access

Advanced Skills (9)

Skill	Purpose	Key Capability
functions	Serverless deployment	Deploy browser automation to Browserbase cloud as serverless functions using bb functions init
site-debugger	Automation debugging	Diagnose failing browser automations by analyzing bot detection, selectors, timing, auth, and CAPTCHAs
browser-trace	Performance tracing	Capture full DevTools-protocol traces (CDP firehose, screenshots, DOM dumps) and bisect them into per-page searchable buckets
cookie-sync	Authentication transfer	Sync cookies from local Chrome to Browserbase persistent contexts so remote sessions can access authenticated sites
autobrowse	Self-improving automation	Iterative loop: inner agent browses, outer agent reads traces and improves the strategy until it reliably passes
ui-test	Adversarial UI testing	AI-powered testing that analyzes git diffs to test changes, or explores the full app to find bugs
company-research	Business intelligence	Research companies by extracting data from their websites and compiling structured reports
event-prospecting	Event discovery	Find and extract event information from conference and meetup platforms

Takeaway: With a single npx skills add browserbase/skills command, Claude Code gains 13 specialized skills covering everything from simple page fetching to self-improving browser automation – no manual configuration needed.

AutoBrowse: The Self-Improving Browser Skill

The AutoBrowse skill is the most innovative component of the Browserbase Skills ecosystem. It implements a self-improving loop where an inner agent browses a website and an outer agent reads the execution traces and improves the navigation strategy until it reliably passes.

How the Loop Works

1. Task Definition – The user provides a task (e.g., “book a flight on Google Flights”) or a URL. AutoBrowse creates a task definition in ./autobrowse/tasks/<task>/task.md that specifies the URL, inputs, steps, and expected JSON output.

2. Inner Agent Execution – The inner agent runs evaluate.mjs which launches a browser session, follows the current strategy, and records every action. It writes a full trace to ./autobrowse/traces/<task>/latest/ including screenshots, accessibility tree snapshots, and a summary with duration, cost, and turn count.

3. Trace Analysis – The outer agent reads the trace summary and identifies exactly where things went wrong. It looks for specific failure patterns: elements not found, timing issues, incorrect selectors, or bot detection triggers.

4. Hypothesis Formation – Based on the trace analysis, the outer agent forms a single, testable hypothesis. For example: “After clicking the dropdown, wait 1 second – options animate in before they are clickable” or “Navigate directly to /pay-invoice/ instead of going through the landing page.”

5. Strategy Update – The outer agent edits strategy.md to incorporate the hypothesis while keeping everything that worked in previous iterations. Good strategies include fast paths (direct URLs), step-by-step workflows with timing notes, site-specific selector knowledge, and failure recovery procedures.

6. Iteration and Graduation – The loop runs for a configurable number of iterations (default: 5). If the task passes on 2 or more of the last 3 iterations, the skill graduates and is installed to ~/.claude/skills/<task-name>/SKILL.md as a permanent, self-contained skill.

Multi-Task Parallel Execution

AutoBrowse also supports running multiple tasks in parallel using sub-agents. Each sub-agent runs the full evaluate-improve loop independently for its assigned task, and the main agent collects results into a structured session report with per-task pass rates, costs, and key learnings.

Important: AutoBrowse’s self-improving loop is a breakthrough in browser automation – the inner agent browses and the outer agent reads traces and improves strategy, creating skills that get better with each iteration until they reliably pass.

Dual Environment: Local and Remote Browser Control

One of the most powerful features of Browserbase Skills is its seamless dual-environment system. The same browse commands work identically whether you are running a local Chrome instance or a remote Browserbase cloud session.

Local Mode

Local mode is the default for development and trusted sites. It offers three variants:

• browse env local – Starts a clean, isolated local browser. Best for reproducible testing and localhost development.
• browse env local --auto-connect – Reuses an already-running debuggable Chrome instance with your existing cookies and login state. Falls back to isolated mode if nothing is available.
• browse env local <port|url> – Attaches to a specific Chrome DevTools Protocol (CDP) target. Useful for connecting to an already-running browser on a specific port.

Local mode is faster, requires no API key, and is ideal for development, localhost testing, and browsing simple sites like documentation wikis and public APIs.

Remote Mode (Browserbase Cloud)

Remote mode activates Browserbase’s cloud infrastructure with a single command: browse env remote. This provides:

• Anti-bot stealth – Custom Chromium build with anti-fingerprinting that bypasses bot detection systems
• Automatic CAPTCHA solving – Handles reCAPTCHA, hCaptcha, and Turnstile challenges automatically
• Residential proxies – Route traffic through residential IPs in 201 countries with geo-targeting
• Session persistence – Cookies and authentication state persist across sessions via Browserbase contexts
• Cookie sync – The cookie-sync skill transfers your local Chrome cookies to a Browserbase context, so remote sessions start already authenticated

When to Switch Environments

The system provides clear guidance on when to use each mode:

Scenario	Recommended Mode	Command
Localhost development	Local (isolated)	browse env local
Reuse local login state	Local (auto-connect)	browse env local --auto-connect
Simple browsing (docs, wikis)	Local (isolated)	browse env local
Protected sites (CAPTCHAs, bot detection)	Remote	browse env remote
Geo-restricted content	Remote	browse env remote
Production scraping	Remote	browse env remote

Important: The dual-environment design means you can develop and test locally with browse env local, then switch to browse env remote for production work – all with the same commands, no code changes required.

Installation

Installing Browserbase Skills takes just one command. The system supports multiple installation methods depending on your preferred workflow.

Quick Install (All Agents)

npx skills add browserbase/skills

Claude Code Plugin Install

On Claude Code, add the marketplace and install the plugin:

/plugin marketplace add browserbase/skills

Then install the browse plugin:

/plugin install browse@browserbase

Manual Install (Claude Code)

1. On Claude Code, type /plugin
2. Select option 3. Add marketplace
3. Enter the marketplace source: browserbase/skills
4. Press enter to select the browse plugin
5. Hit enter again to Install now
6. Restart Claude Code for changes to take effect

Prerequisites

• Node.js 18+ is required for the browse CLI and AutoBrowse
• Chrome/Chromium must be installed for local mode browsing
• Browserbase API key is needed for remote mode (get one at https://browserbase.com/settings)
• ANTHROPIC_API_KEY is required for AutoBrowse’s inner agent

Install the Browse CLI

npm install -g @browserbasehq/browse-cli

Verify Installation

which browse
browse status

Usage Examples

Once installed, you can ask Claude to perform browser tasks using natural language:

• “Go to Hacker News, get the top post comments, and summarize them”
• “QA test http://localhost:3000 and fix any bugs you encounter”
• “Order me a pizza, you’re already signed in on Doordash”
• “Use bb to list my Browserbase projects and show the output as JSON”
• “Initialize a new Browserbase Function with bb functions init and explain the next commands”

Basic Browser Workflow

# Start with local mode for development
browse env local

# Navigate to a page
browse open https://example.com

# Get the accessibility tree (preferred over screenshots)
browse snapshot

# Click an element by its ref from the snapshot
browse click @0-5

# Fill a form field
browse fill "#email" "user@example.com"

# Get the page title
browse get title

# Stop the browser when done
browse stop

Switching to Remote Mode

# Switch to Browserbase cloud for protected sites
browse env remote

# Open a page with an authenticated context
browse open https://protected-site.com --context-id ctx_abc123 --persist

# Take a snapshot
browse snapshot

# Stop the session
browse stop

Key Features

Feature	Description
13 Specialized Skills	Complete coverage from basic browsing to self-improving automation
Dual Environment	Seamless switching between local Chrome and Browserbase cloud
Accessibility-First	Uses accessibility tree snapshots instead of screenshots for faster, more reliable interaction
AutoBrowse Loop	Self-improving browser automation that iterates until skills reliably pass
Anti-Bot Stealth	Custom Chromium with anti-fingerprinting for bypassing bot detection
CAPTCHA Solving	Automatic reCAPTCHA, hCaptcha, and Turnstile solving
Residential Proxies	201 countries with geo-targeting for accessing geo-restricted content
Cookie Sync	Transfer local Chrome cookies to Browserbase contexts for authenticated remote sessions
UI Testing	Adversarial AI-powered testing that tries to break your application
Browser Tracing	Full CDP trace capture with per-page bisection for debugging
Serverless Functions	Deploy browser automation to Browserbase cloud as serverless functions
Daemon Architecture	Persistent browse CLI daemon that auto-starts and manages sessions

Troubleshooting

Chrome not found: Install Chrome for your platform:

• macOS or Windows: https://www.google.com/chrome/
• Linux: sudo apt install google-chrome-stable

Profile refresh: To refresh cookies from your main Chrome profile:

rm -rf .chrome-profile

No active page: Run browse stop, then check browse status. If it still says running, kill the zombie daemon:

pkill -f "browse.*daemon"

Browserbase fails: Verify your API key is set:

echo $BROWSERBASE_API_KEY

Links

• GitHub Repository: https://github.com/browserbase/skills
• Browserbase Website: https://www.browserbase.com

Enjoyed this post? Never miss out on future posts by following us