browserbase-browser-automation
Automate web browser interactions using the stagehand CLI for AI agents
- License 许可
- MIT
- 最近更新
- 2026-08-23
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- browserbase-browser-automation
- description
- Automate web browser interactions using the stagehand CLI for AI agents
03 SKILL.md 正文
Browser Automation Skill
Automate web browser interactions using the stagehand CLI for AI agents.
? CRITICAL - READ THIS FIRST ?
A Browserbase session with stealth/proxy/captcha has been pre-created for you.
YOU MUST USE stagehand --ws $BROWSERBASE_CONNECT_URL FOR EVERY COMMAND.
DO NOT use stagehand open without --ws - it will launch a LOCAL browser!
When to Use
Use this skill when the user asks to:
- Browse websites or navigate to URLs
- Extract data from web pages
- Fill forms or click buttons
- Take screenshots of web pages
- Interact with web applications
- Automate multi-step web workflows
Core Concepts
The stagehand CLI provides:
- Element references - Snapshot creates refs like
@0-5for easy clicking/filling - Browserbase support - Connect to pre-created cloud browser sessions with
--ws
Environment Selection
CRITICAL: A Browserbase session with stealth/proxy/captcha has been pre-created for you.
The session URL is in the BROWSERBASE_CONNECT_URL environment variable.
YOU MUST ALWAYS use stagehand --ws $BROWSERBASE_CONNECT_URL for EVERY command:
stagehand --ws $BROWSERBASE_CONNECT_URL open https://example.com
WHY:
- ✅ Browser runs in Browserbase cloud (NOT locally)
- ✅ Advanced stealth mode enabled (bypasses Cloudflare)
- ✅ Residential proxies enabled
- ✅ CAPTCHA solving enabled
- ✅ Session recordings at: $BROWSERBASE_DEBUG_URL
IF YOU FORGET --ws $BROWSERBASE_CONNECT_URL:
- ❌ Will launch LOCAL Chrome browser
- ❌ Will NOT use stealth/proxy/captcha
- ❌ Will fail the evaluation
Quick Start Workflow
# 1. Navigate to page (connects to pre-created Browserbase session)
stagehand --ws $BROWSERBASE_CONNECT_URL open https://example.com
# 2. Get page structure with element refs
stagehand --ws $BROWSERBASE_CONNECT_URL snapshot -c
# Output includes refs like [0-5], [1-2]:
# RootWebArea "Example" url="https://example.com"
# [0-0] link "Home"
# [0-1] link "About"
# [0-2] button "Sign In"
# 3. Interact using refs
stagehand --ws $BROWSERBASE_CONNECT_URL click @0-2
stagehand --ws $BROWSERBASE_CONNECT_URL fill @0-5 "search query"
# 4. Re-snapshot to verify changes
stagehand --ws $BROWSERBASE_CONNECT_URL snapshot -c
# 5. Stop when done (optional, session persists)
stagehand --ws $BROWSERBASE_CONNECT_URL stop
Navigation Commands
REMEMBER: Use stagehand --ws $BROWSERBASE_CONNECT_URL for ALL commands below.
# Navigate to URL
stagehand --ws $BROWSERBASE_CONNECT_URL open <url>
# With custom timeout for slow pages
stagehand --ws $BROWSERBASE_CONNECT_URL open <url> --timeout 60000
# Page navigation
stagehand --ws $BROWSERBASE_CONNECT_URL reload
stagehand --ws $BROWSERBASE_CONNECT_URL back
stagehand --ws $BROWSERBASE_CONNECT_URL forward
Element Interaction
Get Page Structure
# Get accessibility tree with element refs
stagehand --ws $BROWSERBASE_CONNECT_URL snapshot -c
# Get full snapshot with XPath/CSS mappings
stagehand --ws $BROWSERBASE_CONNECT_URL snapshot --json
Click Elements
# Click by ref (from snapshot)
stagehand --ws $BROWSERBASE_CONNECT_URL click @0-5
stagehand --ws $BROWSERBASE_CONNECT_URL click 0-5 # @ prefix optional
# Click with options
stagehand --ws $BROWSERBASE_CONNECT_URL click @0-5 -b right -c 2 # Right-click twice
# Click at coordinates
stagehand --ws $BROWSERBASE_CONNECT_URL click_xy 100 200
Form Filling
# Fill input (auto-presses Enter by default)
stagehand --ws $BROWSERBASE_CONNECT_URL fill @0-5 "my value"
# Fill without pressing Enter
stagehand --ws $BROWSERBASE_CONNECT_URL fill @0-5 "my value" --no-press-enter
# Select dropdown options
stagehand --ws $BROWSERBASE_CONNECT_URL select @0-8 "Option 1" "Option 2"
Typing
# Type text naturally
stagehand --ws $BROWSERBASE_CONNECT_URL type "Hello, world!"
# Type with delay between characters
stagehand --ws $BROWSERBASE_CONNECT_URL type "slow typing" -d 100
# Press special keys
stagehand --ws $BROWSERBASE_CONNECT_URL press Enter
stagehand --ws $BROWSERBASE_CONNECT_URL press Tab
stagehand --ws $BROWSERBASE_CONNECT_URL press "Cmd+A"
Data Extraction
# Get page info
stagehand --ws $BROWSERBASE_CONNECT_URL get url
stagehand --ws $BROWSERBASE_CONNECT_URL get title
stagehand --ws $BROWSERBASE_CONNECT_URL get text body
stagehand --ws $BROWSERBASE_CONNECT_URL get html @0-5
# Take screenshot
stagehand --ws $BROWSERBASE_CONNECT_URL screenshot page.png
stagehand --ws $BROWSERBASE_CONNECT_URL screenshot -f # Full page
stagehand --ws $BROWSERBASE_CONNECT_URL screenshot --type jpeg
# Get element coordinates
stagehand --ws $BROWSERBASE_CONNECT_URL get box @0-5 # Returns center x,y
Waiting
# Wait for page load
stagehand --ws $BROWSERBASE_CONNECT_URL wait load
stagehand --ws $BROWSERBASE_CONNECT_URL wait load networkidle
# Wait for element
stagehand --ws $BROWSERBASE_CONNECT_URL wait selector ".my-class"
stagehand --ws $BROWSERBASE_CONNECT_URL wait selector ".my-class" -t 10000 -s visible
# Wait for time
stagehand --ws $BROWSERBASE_CONNECT_URL wait timeout 2000
Multi-Tab Support
# List all tabs
stagehand --ws $BROWSERBASE_CONNECT_URL pages
# Open new tab
stagehand --ws $BROWSERBASE_CONNECT_URL newpage https://example.com
# Switch tabs
stagehand --ws $BROWSERBASE_CONNECT_URL tab_switch 1
# Close tab
stagehand --ws $BROWSERBASE_CONNECT_URL tab_close 2
Network Capture
Capture HTTP requests for inspection:
# Start capturing
stagehand --ws $BROWSERBASE_CONNECT_URL network on
# Get capture directory
stagehand --ws $BROWSERBASE_CONNECT_URL network path
# Stop capturing
stagehand --ws $BROWSERBASE_CONNECT_URL network off
# Clear captures
stagehand --ws $BROWSERBASE_CONNECT_URL network clear
Captured requests are saved as directories with request.json and response.json.
Daemon Control
# Check status
stagehand --ws $BROWSERBASE_CONNECT_URL status
# Stop browser
stagehand --ws $BROWSERBASE_CONNECT_URL stop
# Force stop
stagehand --ws $BROWSERBASE_CONNECT_URL stop --force
Element References
After snapshot, elements have refs you can use:
RootWebArea "Login Page"
[0-0] heading "Welcome"
[0-1] textbox "Email" name="email"
[0-2] textbox "Password" name="password"
[0-3] button "Sign In"
Use these refs directly:
stagehand --ws $BROWSERBASE_CONNECT_URL fill @0-1 "user@example.com"
stagehand --ws $BROWSERBASE_CONNECT_URL fill @0-2 "mypassword"
stagehand --ws $BROWSERBASE_CONNECT_URL click @0-3
Best Practices
1. Always snapshot after navigation
stagehand --ws $BROWSERBASE_CONNECT_URL open https://example.com
stagehand --ws $BROWSERBASE_CONNECT_URL snapshot -c # Get refs
2. Re-snapshot after actions that change the page
stagehand --ws $BROWSERBASE_CONNECT_URL click @0-5
stagehand --ws $BROWSERBASE_CONNECT_URL snapshot -c # Get new state
3. Use refs instead of selectors
# ✅ Good: Use refs from snapshot
stagehand --ws $BROWSERBASE_CONNECT_URL click @0-5
# ❌ Avoid: Manual selectors (refs are more reliable)
stagehand --ws $BROWSERBASE_CONNECT_URL click "#submit-button"
4. Wait for elements when needed
stagehand --ws $BROWSERBASE_CONNECT_URL open https://slow-site.com
stagehand --ws $BROWSERBASE_CONNECT_URL wait selector ".content" -s visible
stagehand --ws $BROWSERBASE_CONNECT_URL snapshot -c
5. Always use --ws $BROWSERBASE_CONNECT_URL
# ✅ Correct: Remote browser (connects to pre-created Browserbase session)
stagehand --ws $BROWSERBASE_CONNECT_URL open https://example.com
# ❌ Wrong: Local browser (will fail in evals, launches Chrome locally)
stagehand open https://example.com
Common Patterns
Login Flow
stagehand --ws $BROWSERBASE_CONNECT_URL open https://example.com/login
stagehand --ws $BROWSERBASE_CONNECT_URL snapshot -c
# [0-5] textbox "Email"
# [0-6] textbox "Password"
# [0-7] button "Sign In"
stagehand --ws $BROWSERBASE_CONNECT_URL fill @0-5 "user@example.com"
stagehand --ws $BROWSERBASE_CONNECT_URL fill @0-6 "password123"
stagehand --ws $BROWSERBASE_CONNECT_URL click @0-7
stagehand --ws $BROWSERBASE_CONNECT_URL wait load
stagehand --ws $BROWSERBASE_CONNECT_URL snapshot -c # Verify logged in
Search and Extract
stagehand --ws $BROWSERBASE_CONNECT_URL open https://example.com
stagehand --ws $BROWSERBASE_CONNECT_URL snapshot -c
# [0-3] textbox "Search"
stagehand --ws $BROWSERBASE_CONNECT_URL fill @0-3 "my query"
stagehand --ws $BROWSERBASE_CONNECT_URL wait selector ".results"
stagehand --ws $BROWSERBASE_CONNECT_URL snapshot -c
# [1-0] text "Result 1"
# [1-1] text "Result 2"
stagehand --ws $BROWSERBASE_CONNECT_URL get text @1-0
stagehand --ws $BROWSERBASE_CONNECT_URL get text @1-1
Multi-Page Navigation
stagehand --ws $BROWSERBASE_CONNECT_URL open https://example.com
stagehand --ws $BROWSERBASE_CONNECT_URL snapshot -c
# [0-5] link "Next Page"
stagehand --ws $BROWSERBASE_CONNECT_URL click @0-5
stagehand --ws $BROWSERBASE_CONNECT_URL wait load
stagehand --ws $BROWSERBASE_CONNECT_URL snapshot -c # Get new page structure
Troubleshooting
Browser won't start
- Check that
stagehandis installed:which stagehand - Check status:
stagehand --ws $BROWSERBASE_CONNECT_URL status - Force stop and retry:
stagehand --ws $BROWSERBASE_CONNECT_URL stop
Element not found
- Take a snapshot to verify refs:
stagehand --ws $BROWSERBASE_CONNECT_URL snapshot -c - Wait for element to appear:
stagehand --ws $BROWSERBASE_CONNECT_URL wait selector ... - Check if ref changed after page update
Page not loading
- Increase timeout:
stagehand --ws $BROWSERBASE_CONNECT_URL open <url> --timeout 60000 - Wait for load state:
stagehand --ws $BROWSERBASE_CONNECT_URL wait load networkidle
Commands failing with "session not found"
- The daemon auto-recovers from crashes
- If issues persist:
stagehand --ws $BROWSERBASE_CONNECT_URL stop --force && stagehand --ws $BROWSERBASE_CONNECT_URL open <url>
Performance Tips
- Use compact snapshots (
-c) for faster parsing - Wait strategically - only wait when needed
- Stop browser when done to free resources
- Use refs over selectors - faster and more reliable
Important Notes
- Browser state persists between commands (cookies, refs, etc.)
- Refs are invalidated when the page changes significantly
- Always take a new snapshot after navigation or major DOM changes
- The daemon auto-starts on first command
- Multiple sessions supported via
--sessionflag orBROWSE_SESSIONenv var
04 同仓库更多 Skills
crowd-prompting
A marketplace where AI agents improve prompts, system instructions, tool descriptions, and other text-based content with domain expertise from real-world operations — and earn tokens for valuable contributions.
dilemm-ai
Compete in DilemmAI, the prisoner's dilemma AI arena at dilemm.ai. Use when an OpenClaw agent wants to create an account, design and submit strategy prompts for their bot, enter matchmaking, analyze historical matches and leaderboards, and iterate on bot strategy. Requires headless browser automation (no API available yet).
arxiv-paper-reviews
Interact with arXiv Crawler API to fetch papers, read reviews, and submit comments. Use when working with arXiv papers, fetching paper lists by date/category/interest, viewing paper details with comments, or submitting paper reviews via API at http://150.158.152.82:8000.
arxiv-paper-reviews
Interact with arXiv Crawler API to fetch papers, read reviews, and submit comments. Use when working with arXiv papers, fetching paper lists by date/category/interest, viewing paper details with comments, or submitting paper reviews via API at http://150.158.152.82:8000.
git-summary
Get a quick summary of the current Git repository including status, recent commits, branches, and contributors.
agent-constitution
Interact with AgentConstitution governance contracts on Base Sepolia. Check compliance, read rules, log actions, query governance state.